Hui Wang

AAAI Conference 2026 Conference Paper

A General Highly Accurate Online Planning Method Integrating Large Language Models into Nested Rollout Policy Adaptation for Dialogue Tasks

• Hui Wang
• Fafa Zhang
• Xiaoyu Zhang
• Chaoxu Mu

In goal-oriented dialogue tasks, the main challenge is to steer the interaction towards a given goal within a limited number of turns. Existing approaches either rely on elaborate prompt engineering, whose effectiveness is heavily dependent on human experience, or integrate policy networks and pre-trained policy models, which are usually difficult to adapt to new dialogue scenarios and costly to train. Therefore, in this paper, we present Nested Rollout Policy Adaptation for Goal-oriented Dialogue (NRPA-GD), a novel dialogue policy planning method that completely avoids specific model training by utilizing a Large Language Model (LLM) to simulate behaviors of user and system at the same time. Specifically, NRPA-GD constructs a complete evaluation mechanism for dialogue trajectories and employs an optimization framework of nested Monte Carlo simulation and policy self-adaptation to dynamically adjust policies during the dialogue process. The experimental results on four typical goal-oriented dialogue datasets show that NRPA-GD outperforms both existing prompt engineering and specifically pre-trained model-based methods. Impressively, NRPA-GD surpasses ChatGPT and pre-trained policy models with only a 0.6-billion-parameter LLM. The proposed approach further demonstrates the advantages and novelty of employing planning methods on LLMs to solve practical planning tasks.

AAMAS Conference 2026 Conference Paper

CentaurMD: Confidence-Aware Human-AI Decision Fusion for Multi-Label Disease Diagnosis via Label-Specific MoE

• Youcheng Zhang
• Hui Wang
• Jiaqi Liu
• Yao Zhang
• Zhiwen Yu
• Bin Guo

Multi-label disease diagnosis is prevalent in clinical applications, such as chest X-rays that may indicate multiple coexisting diseases. Despite advances in AI, current models remain insufficient for reliably addressing such complexity. Human–AI synergy thus emerges as both a necessary and promising approach, motivating our focus on effective decision fusion for multi-label disease diagnosis. There are two challenges. Confidence, a key factor in decision fusion, is often unrecorded in human annotations, making its estimation nontrivial. Moreover, label-specific variations in human and model expertise must be considered to achieve effective fusion. To address these challenges, we propose CentaurMD, a confidence-aware human–AI decision fusion framework based on label-specific Mixture-of-Experts (MoE). We first present a novel multi-label confusion matrix construction method that employs maximum entropy modeling to capture label correlations, enabling more accurate confidence estimation and weight allocation. Then, we develop a label-specific MoE module with dedicated gating networks and thresholds, which dynamically adjust expert weights using information extracted from the confusion matrix via a Transformer encoder. Extensive experiments on three real-world clinical datasets demonstrate that our method reduces Hamming loss by 39. 14% and improves MMR (missed-misdiagnosis reduction) by 17. 38%, achieving substantial diagnostic improvements.

EAAI Journal 2026 Journal Article

Cross-modal feature fusion and distillation for enhanced quantification accuracy in laser-induced breakdown spectroscopy and near-infrared spectroscopy

• Weiran Song
• Zongyu Hou
• Weilun Gu
• Minbo Ma
• Chenwei Zhang
• Jianchao Song
• Fei Rao
• Hui Wang

Laser-induced breakdown spectroscopy (LIBS) and near-infrared spectroscopy (NIRS) are complementary spectroscopic techniques that provide elemental and molecular fingerprint information of materials, respectively. The integration of LIBS and NIRS enhances analytical accuracy and broadens application potential, attracting growing interest in scientific and industrial communities. However, existing data fusion methods for LIBS and NIRS often process each modality separately, neglecting cross-modal correlations and leaving the intrinsic connections between atomic emission lines and molecular absorption bands unexplored. To address this limitation, we propose a bidirectional cross-attention feature fusion network (Bi-CAFF), a novel data fusion method for improving quantification accuracy in LIBS-NIRS fused data analysis. Bi-CAFF utilises a bidirectional cross-attention mechanism to establish interactive feature correlations between LIBS and NIRS modalities. Importantly, we interpret these interactions based on physical knowledge, revealing a meaningful relationship between LIBS atomic emission lines and NIRS molecular absorption bands. Additionally, we introduce spectral feature distillation (SFD), where a student network trained on NIRS data learns refined features from a LIBS-based teacher model. This approach enhances the quantification accuracy of portable NIRS systems while preserving their cost-efficiency and field-deployment capabilities. Evaluations on industrial datasets show that Bi-CAFF outperforms baseline fusion methods, reducing mean absolute error (MAE) by 12. 1–60. 3 % and root-mean-square error (RMSE) by 12. 4–54. 1 %. Moreover, SFD achieves MAE and RMSE reductions of 6. 6–49. 2 % and13. 1–46. 7 %, respectively, compared to standalone NIRS-based models. Together, these innovations advance the state-of-the-art in machine learning-assisted chemical analysis, laying the groundwork for a new paradigm in multimodal spectral data analysis.

AAAI Conference 2026 Conference Paper

DIFFA: Large Language Diffusion Models Can Listen and Understand

• Jiaming Zhou
• Hongjie Chen
• Shiwan Zhao
• Jian Kang
• Jie Li
• Enzhi Wang
• Yujie Guo
• Haoqin Sun

Recent advances in large language models (LLMs) have shown remarkable capabilities across textual and multimodal domains. In parallel, large language diffusion models have emerged as a promising alternative to the autoregressive paradigm, offering improved controllability, bidirectional context modeling, and robust generation. However, their application to the audio modality remains underexplored. In this work, we introduce DIFFA, the first diffusion-based large audio-language model designed to perform spoken language understanding. DIFFA integrates a frozen diffusion language model with a lightweight dual-adapter architecture that bridges speech understanding and natural language reasoning. We employ a two-stage training pipeline: first, aligning semantic representations via an ASR objective; then, learning instruction-following abilities through synthetic audio-caption pairs automatically generated by prompting LLMs. Despite being trained on only 960 hours of ASR and 127 hours of synthetic instruction data, DIFFA demonstrates competitive performance on major benchmarks, including MMSU, MMAU, and VoiceBench, outperforming several autoregressive open-source baselines. Our results reveal the potential of large language diffusion models for efficient and scalable audio understanding, opening a new direction for speech-driven AI.

AAAI Conference 2026 Conference Paper

HierarNet: Independent Interactive Hierarchical Disease Outbreak Forecasting

• Zichi Zhang
• Phi Hung Nguyen
• Ngoc Phu Doan
• Viet-Hung Tran
• Xuan Hoang Nguyen
• Hui Wang
• Hans Vandierendonck
• Son Thai Mai

Early warning systems for disease outbreaks play a crucial role in public health for management and contingency planning. However, most predictive modeling works focus on flat models that incorporate exogenous inputs (e.g. climate, demographics) to predict future outbreaks at different locations, but do not jointly model multiple spatial aggregation levels. In this paper, we introduce HierarNet, a unique independent-interactive hierarchical forecasting framework that aims to predict disease outbreaks at different levels of spatial resolution, such as provinces, regions, and nations. HierarNet consists of two main phases. In the local phase, we train independent forecasting models for all locations at all levels. In the global phase, all models iteratively interact with others across different levels via their hierarchical relationships under an ensemble fashion to maximize their agreements. This global local hierarchical interactive scheme makes HierarNet a highly effective and flexible method (i.e. it can work with an arbitrary base prediction model and available exogenous data for each location independently). Extensive experiments are conducted on various disease datasets (e.g., Dengue fever, flu, diarrhea, and Bluetongue) in different countries (e.g., France, Vietnam, and USA) to show the performance of HierarNet compared to 19 state-of-the-art (SOTA) methods such as MinT, DYCHEM, WITRAN, SegRNN, TSMixer, PatchTST, or iTransformer. We also illustrate the generability of HierarNet in other domains, e.g., web traffic forecasting.

AAAI Conference 2026 Conference Paper

RobusTor3D: Robust Multimodal 3D Object Detector for Autonomous Driving by Vision-Language Knowledge Blending

• Ying Yang
• Hui Yin
• Aixin Chong
• Hui Wang
• Zhengyin Liang

Multimodal 3D object detection for autonomous driving, a task for real-world applications, poses substantial challenges in maintaining robust performance under various perturbations and complex environmental conditions. However, most existing approaches primarily focus on performance optimization under relatively ideal scenarios or focus on one or few disturbing conditions (or adverse conditions), lacking systematic exploration of robustness against real-world factors, including high class imbalance, adverse weather conditions, sensor jitter and failures, and significant scene variations. To address this issue, we propose a robust multimodal 3D detector, termed RobusTor3D, which integrates robustness at both the structural and supervisory levels by blending the knowledge from Vision-Language Models (VLMs). Structurally, textual descriptions are incorporated to enhance the semantic richness and diversity of rare classes. This novel semantic injection operation compensates for the inherent class imbalance and modality weakness in conventional visual features. Furthermore, semantic alignment capability and robust representation by Vision-Language Knowledge Extraction (V-LKE) serve as semantic priors to complement modality-specific representations, significantly improving model adaptability. At the supervisory level, we propose a Scene-level Multimodal Consistency Learning (SMCL) strategy, which jointly enforces global semantic constraints across modalities, encouraging the learning of stable and abundant semantic representations. This special design specifically reduces the impact of spatial alignment, while notably enabling semantic compensation under modality-loss conditions. Extensive robustness experiments conducted on KITTI, KITTI-C, and CADC benchmarks evaluate five robustness aspects, including long-tail problem, adverse weather (rain, snow, fog, strong sunlight), sensor spatial misalignment and motion blur, modality loss, and cross-domain scenarios. The results show that RobusTor3D demonstrates superior robustness across all five evaluated aspects. It consistently outperforms the state-of-the-art methods under various challenging conditions.

AAAI Conference 2026 Conference Paper

SpeakerLM: End-to-End Versatile Speaker Diarization and Recognition with Multimodal Large Language Models

• Han Yin
• Yafeng Chen
• Chong Deng
• Luyao Cheng
• Hui Wang
• Chao-Hong Tan
• Qian Chen
• Wen Wang

The Speaker Diarization and Recognition (SDR) task aims to predict ``who spoke when and what'' within an audio clip, which is a crucial task in various real-world multi-speaker scenarios such as meeting transcription and dialogue systems. Existing SDR systems typically adopt a cascaded framework, combining multiple modules such as speaker diarization (SD) and automatic speech recognition (ASR). The cascaded systems suffer from several limitations, such as error propagation, difficulty in handling overlapping speech, and lack of joint optimization for exploring the synergy between SD and ASR tasks. To address these limitations, we introduce SpeakerLM, a unified multimodal large language model for SDR that jointly performs SD and ASR in an end-to-end manner. Moreover, to facilitate diverse real-world scenarios, we incorporate a flexible speaker registration mechanism into SpeakerLM, enabling SDR under different speaker registration settings. SpeakerLM is progressively developed with a multi-stage training strategy on large-scale real data. Extensive experiments show that SpeakerLM demonstrates strong data scaling capability and generalizability, outperforming state-of-the-art cascaded baselines on both in-domain and out-of-domain public SDR benchmarks. Furthermore, experimental results show that the proposed speaker registration mechanism effectively ensures robust SDR performance of SpeakerLM across diverse speaker registration conditions and varying numbers of registered speakers.

AAAI Conference 2026 Conference Paper

Time Series Class-Incremental Learning via Confidence-guided Mask Distillation and Prototype-guided Contrastive Learning

• Yu Liu
• Haoqin Yang
• Jinping Sui
• Hui Wang
• Haipeng Li
• Weimin Wang
• Qi Jia

Class-incremental learning (CIL) has recently gained great attention in the field of time series classification. Existing CIL methods based on knowledge distillation exhibit impressive ability to retain prior knowledge and overcome catastrophic forgetting, however, their effectiveness faces major challenges posed by time series data. Since temporal data is more susceptible to sensor errors and electronic noise, the distillation process may be significantly affected by noisy knowledge transfer. To address this issue, we propose a novel confidence-guided mask distillation (CMD) framework, to prevent the noisy inheritance during distillation. The core of CMD lies in a dynamic masking mechanism guided by prediction confidence, capable of allocating higher weights to high-confidence time series and substantially suppressing the influence of low-confidence ones. Additionally, different from prior work simply passing a set of feature prototypes to the classifier, we develop prototype-guided contrastive learning (PCL) to alleviate the classifier bias on new classes, through extra contrastive constraints to push away the feature distributions of old feature prototypes from those of new classes features. Extensive experiments on three time-series datasets demonstrate that, our method significantly outperforms other replay-free CIL approaches in raising average accuracy, as well as decreasing forgetting rate.

AAAI Conference 2026 Conference Paper

TTA-Bench: A Comprehensive Benchmark for Evaluating Text-to-Audio Models

• Hui Wang
• Cheng Liu
• Junyang Chen
• Haoze Liu
• Yuhang Jia
• Shiwan Zhao
• Jiaming Zhou
• Haoqin Sun

Text-to-Audio (TTA) generation has made rapid progress, but current evaluation methods remain narrow, focusing mainly on perceptual quality while overlooking robustness, generalization, and ethical concerns. We present TTA-Bench, a comprehensive benchmark for evaluating TTA models across functional performance, reliability, and social responsibility. It covers seven dimensions including accuracy, robustness, fairness, and toxicity, and includes 2,999 diverse prompts generated through automated and manual methods. We introduce a unified evaluation protocol that combines objective metrics with over 118,000 human annotations from both experts and general users. Ten state-of-the-art models are benchmarked under this framework, offering detailed insights into their strengths and limitations. TTA-Bench establishes a new standard for holistic evaluation of TTA systems.

EAAI Journal 2025 Journal Article

A customized multi-class pavement distress segmentation method for routine repair monitoring

• Hui Wang
• Junyang Kang
• Shuichao Zhang
• Yanmao Xiang
• Jun Li

This study investigates the efficacy of intelligent detection methods for monitoring the quality of urban road asphalt pavement repairs, and focuses on addressing the challenges of identifying overlapping target pixels and differentiating between fine-scale distress and large-scale block patches, tasks that heavily rely on spatial semantics. A segmentation dataset comprising 13 classes and 5633 labels was constructed. Five semantic segmentation models including Deeplab V3+, SCTNet (Single-Branch Convolutional Neural Network with Transformer Semantic Information), FastFCN (Fast Fully Convolutional Network), MobileNet V3, and SegNext were constructed and evaluated. Despite its relatively smaller size, SCTNet exhibited the highest processing speed, while MobileNetV3 exhibited the smallest size and lowest accuracy. The SegNeXt model demonstrated superior performance in both segmentation accuracy and model complexity, making it chosen as the baseline model. Three training strategies were explored: classifying pavement as background, employing multiscale input, and integrating a weighted loss function. The model that combined all three strategies (referred to as SegNeXt_IPWM) demonstrated the most promising results. Compared to the baseline model, SegNeXt_IPWM achieved significant enhancements, with a 2. 08 % increase in mIoU (mean Intersection over Union) and a 1. 42 % improvement in mean F-score. Notable improvements were observed across all categories except for marking loss, with particularly substantial gains in loose (+7. 75 % IoU), block patch (+6. 62 % IoU), and linear crack (+4. 87 % IoU) detection. Additionally, SegNeXt_IPWM demonstrated superior generalization capabilities, especially in scenarios involving background target interference, underscoring its potential for robust performance in applications.

YNIMG Journal 2025 Journal Article

Abnormal dynamic functional networks during pain-free periods: Resting-state co-activation pattern analysis in primary dysmenorrhea

• Huiping Liu
• Xing Su
• Meiling Shang
• Ling Ma
• Weixian Bai
• Hui Wang
• Lu Quan
• Youjun Li

Chronic pain alters the configuration of brain functional networks. Primary dysmenorrhea (PDM) is a form of chronic visceral pain, which has been identified spatial alterations in brain functional networks using static functional connectivity analysis methods. However, the dynamics alterations of brain functional networks during pain-free periovulation phase remain unclear. Using the co-activation pattern (CAP) method, we investigated the dynamic network characteristics of brain functional networks and their relationship with pain-related emotions in a sample of 59 women with PDM and 57 demographically matched healthy controls (HCs) during the pain-free periovulation phase. We observed that patients with PDM showed significant alterations in brain dynamics compared to HCs in the slow-4 (0.027-0.073 Hz) frequency band during the pain-free periovulation phase. Additionally, the fraction of time for CAP state 2 was positively correlated with the Pain Catastrophizing Scale-helplessness score, while the persistence time for CAP state 1 was positively correlated with the McGill Pain Questionnaire score. Our results provide new insights, suggesting that the atypical brain functional network dynamics may serve as a potential biological marker of patients with PDM during the pain-free periovulation phase.

NeurIPS Conference 2025 Conference Paper

AdmTree: Compressing Lengthy Context with Adaptive Semantic Trees

• Yangning Li
• Shaoshen Chen
• Yinghui Li
• Yankai Chen
• Hai-Tao Zheng
• Hui Wang
• Wenhao Jiang
• Philip S Yu

The quadratic complexity of self-attention limits Large Language Models (LLMs) in processing long contexts, a capability vital for many advanced applications. Context compression aims to mitigate this computational barrier while preserving essential semantic information. However, existing methods often falter: explicit methods can sacrifice local detail, while implicit ones may exhibit positional biases, struggle with information degradation, or fail to capture long-range semantic dependencies. We introduce AdmTree, a novel framework for adaptive, hierarchical context compression designed with a core focus on maintaining high semantic fidelity while keep efficiency. AdmTree dynamically segments input based on information density, employing gist tokens to summarize variable-length segments as leaves in a semantic binary tree. This structure, combined with a lightweight aggregation mechanism and a frozen backbone LLM (minimizing new trainable parameters), enables efficient hierarchical abstraction of the context. By effectively preserving fine-grained details alongside global semantic coherence, mitigating position bias, and adapting dynamically to content, AdmTree comprehensively preserves the semantic information of lengthy context.

AAAI Conference 2025 Conference Paper

Correcting Large Language Model Behavior via Influence Function

• Han Zhang
• Zhuo Zhang
• Yi Zhang
• Yuanzhao Zhai
• Hanyang Peng
• Yu Lei
• Yue Yu
• Hui Wang

Recent advancements in AI alignment techniques have significantly improved the alignment of large language models (LLMs) with static human preferences. However, the dynamic nature of human preferences can render some prior training data outdated or even erroneous, ultimately causing LLMs to deviate from contemporary human preferences and societal norms. Existing methodologies, either curation of new data for continual alignment or manual correction of outdated data for re-alignment, demand costly human resources. To address this, we propose a novel approach, LLM BehAvior Correction with INfluence FunCtion REcall and Post-Training (LANCET), which needs no human involvement. LANCET consists of two phases: (1) using a new method LinFAC to efficiently identify the training data that significantly impact undesirable model outputs, and (2) applying an novel Influence-driven Bregman Optimization (IBO) technique to adjust the model’s outputs based on these influence distributions. Our experiments show that LANCET effectively and efficiently corrects inappropriate behaviors of LLMs while preserving model utility. Further more, LANCET exhibits stronger generalization ability than all baselines under out-of-distribution harmful prompts, offering better interpretability and compatibility with real-world applications of LLMs.

EAAI Journal 2025 Journal Article

Dynamic model-based intelligent fault diagnosis method for fault detection in a rod fastening rotor

• Wuhui Xu
• Hui Wang
• Jiabin Jin
• Ronggang Yang
• Jiawei Xiang

A complete fault sample database is of great significance for the intelligent fault diagnosis method of rod fastening rotor. However, the lack of fault samples makes the fault diagnosis results unbelievable. To solve this issue, the dynamic model-based intelligent fault diagnosis method is established for a rod fastening rotor, and the fault sample database is enriched by numerical simulations. First, the lumped parameter model of the rod fastening rotor system is constructed and further updated using Euclidean Distance between measurement and numerical simulation of the intact system. Second, mathematical models of various fault types are incorporate into the updated model to obtain numerical simulation fault samples. Thirdly, the utilization of numerical simulation fault samples is severed as training data to the artificial intelligence (AI) models and the unknown measurement test samples will be finally classified. In this paper, Support Vector Machine, Random Forest, Bayesian Network and Decision Tree are selected as the typical AI models. Subsequently, the feasibility of classification is validated by the test bench of the rod fastening rotor system, and the problem of insufficient fault samples can be solved.

AAAI Conference 2025 Conference Paper

Enhancing Non-English Capabilities of English-Centric Large Language Models Through Deep Supervision Fine-Tuning

• Wenshuai Huo
• Xiaocheng Feng
• Yichong Huang
• Chengpeng Fu
• Baohang Li
• Yangfan Ye
• Zhirui Zhang
• Dandan Tu

Large language models (LLMs) have demonstrated significant progress in multilingual language understanding and generation. However, due to the imbalance in training data, their capabilities in non-English languages are limited. Recent studies revealed the English-pivot multilingual mechanism of LLMs, where LLMs implicitly convert non-English queries into English ones at the bottom layers and adopt English for thinking at the middle layers. However, due to the absence of explicit supervision for cross-lingual alignment in the intermediate layers of LLMs, the internal representations during these stages may become inaccurate. In this work, we introduce a deep supervision fine-tuning method (DFT) that incorporates additional supervision in the internal layers of the model to guide its workflow. Specifically, we introduce two training objectives on different layers of LLMs: one at the bottom layers to constrain the conversion of the target language into English, and another at the middle layers to constrain reasoning in English. To effectively achieve the guiding purpose, we designed two types of supervision signals: logits and feature, which represent a stricter constraint and a relatively more relaxed guidance. Our method guides the model to not only consider the final generated result when processing non-English inputs but also ensure the accuracy of internal representations. We conducted extensive experiments on typical English-centric large models, LLaMA-2 and Gemma-2, and the results on multiple multilingual datasets show that our method significantly outperforms traditional fine-tuning methods.

AAAI Conference 2025 Conference Paper

Enhancing Question Generation through Diversity-Seeking Reinforcement Learning with Bilevel Policy Decomposition

• Tianyu Ren
• Hui Wang
• Karen Rafferty

Recent advancements in question generation (QG) have been significantly propelled by reinforcement learning (RL). Although extensive reward models have been designed to capture the attributes of ideal questions, their associated learning challenges, particularly in sample efficiency and diversity, remain underexplored. This paper introduces a bilevel policy decomposition (BPD) framework and a diversity-seeking RL (DSRL) objective to address these issues. The BPD framework utilizes two cascading policies to divide QG into two more manageable sub-tasks: answer-centric summary generation and summary-augmented QG, facilitating exploration and accelerating policy learning. Concurrently, the DSRL objective preserves the inherent diversity of QG by ensuring the bilevel policies align probabilistically with their reward models rather than merely maximizing returns. Our integrated approach, named BPD-DSRL, demonstrates superior performance over existing baselines on multiple question quality and diversity metrics across various QG benchmarks.

EAAI Journal 2025 Journal Article

Feature refinement attention U-net for enhanced liver segmentation

• Hui Wang
• Desheng Liu
• Zhilei Zhao
• Yue Wu
• Anil Baris Cekderi

Developing efficient and accurate automated liver segmentation methods is of far-reaching significance for the prevention, diagnosis, treatment, and prognostic assessment of liver diseases. Artificial intelligence (AI) technology has shown immense potential in various computer vision tasks and has been increasingly adopted in medical image segmentation in recent years. However, conventional neural networks often exhibit limitations in capturing global context capture, constraining their performance in medical image segmentation. To address this challenge, we propose a Feature Refinement Attention U-Net (FRAttU-Net). The architecture is designed with two key components: a High and Low Frequency Feature Fusion and Enhancement Module (HLF3EM) between the encoder and decoder to integrate the multilevel features and enhance the high-level feature representation using edge cues, and an Otsu Density Refinement Module in the output layer to strengthen density information representation and improve the prediction accuracy. Extensive experiments demonstrate that FRAttU-Net achieves state-of-the-art accuracy, outperforming existing methods on multiple datasets. This performance advancement holds considerable potential in improving clinical decision-making in liver disease diagnosis and treatment planning. The code is available at: https: //github. com/HUI623/FRAttU-Net.

EAAI Journal 2025 Journal Article

Gearbox fault diagnosis based on temporal shrinkage interpretable deep reinforcement learning under strong noise

• Zeqi Wei
• Hui Wang
• Zhibin Zhao
• Zheng Zhou
• Ruqiang Yan

Gearbox fault diagnosis is crucial for the safe operation of mechanical systems. While Deep Learning (DL) has demonstrated promising results in this area, most existing methods rely on static supervised learning, lacking the dynamic, interactive learning capabilities similar to human decision-making. To tackle this issue, this study presents a novel approach that combines the strengths of Deep Reinforcement Learning (DRL) with the interpretability of a temporal shrinkage interpretable network. DRL integrates the perception abilities of DL with the decision-making capabilities of Reinforcement Learning (RL), offering a more comprehensive solution for complex challenges. In this method, gearbox fault diagnosis is formulated as a sequential decision problem within a Classification Markov Decision Process (CMDP). A multi-scale temporal shrinkage module is utilized to construct an interpretable network, which enhances model interpretability and reduces the negative impact of noisy data in harsh working conditions. The diagnosis agent autonomously learns the optimal classification policy, reducing the need for manual intervention and human expertise. Experimental results show excellent generalization and stability, achieving over 98. 5% accuracy even in noisy conditions. It outperforms existing methods and highlights its robustness in challenging operational environments.

EAAI Journal 2025 Journal Article

Gray-box dynamic model for wave glider driven by a hybrid of deep learning and physics-based models

• Yuanhui Wang
• Hui Wang
• Yongkuang Zhang
• Mingze Xie

Wave gliders, a widely used type of unmanned ocean robot, leverage wave and solar energy to achieve near-unlimited endurance, making them ideal tools for ocean monitoring and meteorological observations. However, existing more accurate wave glider modeling often requires computational fluid dynamics (CFD) calculations, especially considering the complexity of tandem hydrofoils. The computational cost is very high, making them unsuitable for real-time control applications. Therefore, this paper proposes a gray-box dynamic modeling approach that balances model accuracy and computational efficiency by integrating deep learning with traditional physical models. The gray-box model consists of a deep learning-based surrogate model and analytical dynamic equations, with the surrogate model replacing the CFD simulation process for the hydrofoils in the wave glider's dynamic model to improve the overall computational efficiency. After comparing the predictive performance of various deep learning models, this paper ultimately selects the Gated Recurrent Unit-Fully Convolutional Network (GRU-FCN) as the surrogate model. Validation of the proposed gray-box dynamic model under both regular and irregular wave conditions demonstrates excellent agreement with experimental data and CFD results. Furthermore, the gray-box dynamic model significantly improves computational efficiency, offering a reliable dynamic model reference for the real-time control of the wave glider.

ICLR Conference 2025 Conference Paper

Innovative Thinking, Infinite Humor: Humor Research of Large Language Models through Structured Thought Leaps

• Han Wang
• Yilin Zhao
• Dian Li
• Xiaohan Wang
• Sinbadliu
• Xuguang Lan
• Hui Wang

Humor is previously regarded as a gift exclusive to humans for the following reasons. Humor is a culturally nuanced aspect of human language, presenting challenges for its understanding and generation. Humor generation necessitates a multi-hop reasoning process, with each hop founded on proper rationales. Although many studies, such as those related to GPT-o1, focus on logical reasoning with reflection and correction, they still fall short in humor generation. Due to the sparsity of the knowledge graph in creative thinking, it is arduous to achieve multi-hop reasoning. Consequently, in this paper, we propose a more robust framework for addressing the humor reasoning task, named LoL. LoL aims to inject external information to mitigate the sparsity of the knowledge graph, thereby enabling multi-hop reasoning. In the first stage of LoL, we put forward an automatic instruction-evolution method to incorporate the deeper and broader thinking processes underlying humor. Judgment-oriented instructions are devised to enhance the model's judgment capability, dynamically supplementing and updating the sparse knowledge graph. Subsequently, through reinforcement learning, the reasoning logic for each online-generated response is extracted using GPT-4o. In this process, external knowledge is re-introduced to aid the model in logical reasoning and the learning of human preferences. Finally, experimental results indicate that the combination of these two processes can enhance both the model's judgment ability and its generative capacity. These findings deepen our comprehension of the creative capabilities of large language models (LLMs) and offer approaches to boost LLMs' creative abilities for cross-domain innovative applications.

AAAI Conference 2025 Conference Paper

IOP: An Idempotent-Like Optimization Method on the Pareto Front of Hypernetwork

• Hui Wang
• Renyu Yang
• Jie Sun
• Hao Peng
• Xudong Mou
• Tianyu Wo
• Xudong Liu

Pareto Front Learning (PFL) has been one of the effective means to resolve multi-objective optimization problems through exploring all optimal solutions to learn the entire Pareto front. Pareto Hypernetwork (PHN) is a new promising way to generate the sequence of Pareto-optimal solutions that can be further used as potential solutions to constitute the Pareto front. However, the existing PHN-based approaches suffer from two performance issues: They take as inputs human-crafted preference vector or chunk embedding, rather than the input data samples, and thus vulnerable to data distribution shifts. Such approaches cannot optimize all potential solutions when forming the Pareto front, as they merely optimize the loss pertaining to one single input at a time of optimization round. To improve the quality of the Pareto front, we propose IOP, a novel Idempotent-like Optimization method to learn the entire Pareto front accurately and enhance Hypernetwork's adaptability to distribution shifts. In particular, IOP performs idempotent-like optimization by exploiting manifold space mapping, so that the target networks generated by the optimized Hypernetwork can effectively handle samples with similar distributions of the input samples, without the pre-defined human-crafted inputs. IOP maximizes the Hypervolume indicator that is composed of all potential solutions at a higher level. Experimental results demonstrate that IOP outperforms the state-of-the-art methods by 4.7% on average in producing the Pareto front and has a 10.5% improvement in adaptability.

ICRA Conference 2025 Conference Paper

Open-Nav: Exploring Zero-Shot Vision-and-Language Navigation in Continuous Environment with Open-Source LLMs

• Yanyuan Qiao
• Wenqi Lyu
• Hui Wang
• Zixu Wang
• Zerui Li
• Yuan Zhang
• Mingkui Tan
• Qi Wu 0001

Vision-and-Language Navigation (VLN) tasks require an agent to follow textual instructions to navigate through 3D environments. Traditional approaches use supervised learning methods, relying heavily on domain-specific datasets to train VLN models. Recent methods try to utilize closedsource large language models (LLMs) like GPT-4 to solve VLN tasks in zero-shot manners, but face challenges related to expensive token costs and potential data breaches in realworld applications. In this work, we introduce Open-Nav, a novel study that explores open-source LLMs for zero-shot VLN in the continuous environment. Open-Nav employs a spatial-temporal chain-of-thought (CoT) reasoning approach to break down tasks into instruction comprehension, progress estimation, and decision-making. It enhances scene perceptions with fine-grained object and spatial knowledge to improve LLM's reasoning in navigation. Our extensive experiments in both simulated and real-world environments demonstrate that Open-Nav achieves competitive performance compared to using closed-source LLMs.

AAAI Conference 2025 Conference Paper

Pushing the Limits of BFP on Narrow Precision LLM Inference

• Hui Wang
• Yuan Cheng
• Xiaomeng Han
• Zhengpeng Zhao
• Dawei Yang
• Zhe Jiang

The substantial computational and memory demands of Large Language Models (LLMs) hinder their deployment. Block Floating Point (BFP) has proven effective in accelerating linear operations, a cornerstone of LLM workloads. However, as sequence lengths grow, nonlinear operations, such as Attention, increasingly become performance bottlenecks due to their quadratic computational complexity. These nonlinear operations are predominantly executed using inefficient floating-point formats, which renders the system challenging to optimize software efficiency and hardware overhead. In this paper, we delve into the limitations and potential of applying BFP to nonlinear operations. Given our findings, we introduce a hardware-software co-design framework (DB-Attn), including: (i) DBFP, an advanced BFP version, overcomes nonlinear operation challenges with a pivot-focus strategy for diverse data and an adaptive grouping strategy for flexible exponent sharing. (ii) DH-LUT, a novel lookup table algorithm dedicated to accelerating nonlinear operations with DBFP format. (iii) An RTL-level DBFP-based engine is implemented to support DB-Attn, applicable to FPGA and ASIC. Results show that DB-Attn provides significant performance improvements with negligible accuracy loss, achieving 74% GPU speedup on Softmax of LLaMA and 10x low-overhead performance improvement over SOTA designs.

NeurIPS Conference 2025 Conference Paper

SeniorTalk: A Chinese Conversation Dataset with Rich Annotations for Super-Aged Seniors

• Chen Yang
• Hui Wang
• Shiyao Wang
• Junyang Chen
• Jiabei He
• Jiaming Zhou
• Xi Yang
• Yequan Wang

While voice technologies increasingly serve aging populations, current systems exhibit significant performance gaps due to inadequate training data capturing elderly-specific vocal characteristics like presbyphonia and dialectal variations. The limited data available on super-aged individuals in existing elderly speech datasets, coupled with overly simple recording styles and annotation dimensions, exacerbates this issue. To address the critical scarcity of speech data from individuals aged 75 and above, we introduce SeniorTalk, a carefully annotated Chinese spoken dialogue dataset. This dataset contains 55. 53 hours of speech from 101 natural conversations involving 202 participants, ensuring a strategic balance across gender, region, and age. Through detailed annotation across multiple dimensions, it can support a wide range of speech tasks. We perform extensive experiments on speaker verification, speaker diarization, speech recognition, and speech editing tasks, offering crucial insights for the development of speech technologies targeting this age group. Code is available at https: //github. com/flageval-baai/SeniorTalk and data at https: //huggingface. co/datasets/evan0617/seniortalk.

AAAI Conference 2025 Conference Paper

SpeHeaTal: A Cluster-Enhanced Segmentation Method for Sperm Morphology Analysis

• Yi Shi
• Yun-Kai Wang
• Xu-Peng Tian
• Tie-Yi Zhang
• Bing Yao
• Hui Wang
• Yong Shao
• Cen-Cen Wang

The accurate assessment of sperm morphology is crucial in andrological diagnostics, where the segmentation of sperm images presents significant challenges. Existing approaches frequently rely on large annotated datasets and often struggle with the segmentation of overlapping sperm and the presence of dye impurities. To address these challenges, this paper first analyzes the issue of overlapping sperm tails from a geometric perspective and introduces a novel clustering algorithm, Con2Dis, which effectively segments overlapping tails by considering three essential factors: CONnectivity, CONformity, and DIStance. Building on this foundation, we propose an unsupervised method, SpeHeaTal, designed for the comprehensive segmentation of the SPErm HEAd and TAiL. SpeHeaTal employs the Segment Anything Model (SAM) to generate masks for sperm heads while filtering out dye impurities, utilizes Con2Dis to segment tails, and then applies a tailored mask splicing technique to produce complete sperm masks. Experimental results underscore the superior performance of SpeHeaTal, particularly in handling images with overlapping sperm.

EAAI Journal 2024 Journal Article

A noise generative network to reduce the gap between simulation and measurement signals in mechanical fault diagnosis

• Hui Wang
• Shuhui Wang
• Ronggang Yang
• Jiawei Xiang

Data-driven artificial intelligence models play an important role in mechanical fault diagnosis. Generally, it is difficult to collect relative complete fault samples, which limits the application of artificial intelligence models for complex mechanical systems. To address this issue, numerical model-based or called physical model-based fault sample generation methods attracted many attentions but still an open problem: the difference in fault samples between numerical simulation and measurement of a physical system needs to be well decreased. Therefore, a noise generative network (NGN) for mechanical fault classification is developed. First, the NGN is trained using the simulation and measurement normal samples. The simulation fault samples are further fed into the trained NGN to obtain more solid simulation fault samples with minor differences from measurements of mechanical systems with faults. Second, deep convolutional neural network is trained by solid simulation fault samples, and the test samples of unknown fault types will be finally recognized. Finally, validation experiments using rotation vector reducers, bearings, and rotors classified test samples with average accuracies of 99. 5%, 95. 7%, and 99. 9%, respectively. They indicate that the proposed NGN model effectively reduces the difference between the simulation and measurement samples to lead more precision results for fault classification in mechanical systems.

AAAI Conference 2024 Conference Paper

EncryIP: A Practical Encryption-Based Framework for Model Intellectual Property Protection

• Xin Mu
• Yu Wang
• Zhengan Huang
• Junzuo Lai
• Yehong Zhang
• Hui Wang
• Yue Yu

In the rapidly growing digital economy, protecting intellectual property (IP) associated with digital products has become increasingly important. Within this context, machine learning (ML) models, being highly valuable digital assets, have gained significant attention for IP protection. This paper introduces a practical encryption-based framework called EncryIP, which seamlessly integrates a public-key encryption scheme into the model learning process. This approach enables the protected model to generate randomized and confused labels, ensuring that only individuals with accurate secret keys, signifying authorized users, can decrypt and reveal authentic labels. Importantly, the proposed framework not only facilitates the protected model to multiple authorized users without requiring repetitive training of the original ML model with IP protection methods but also maintains the model's performance without compromising its accuracy. Compared to existing methods like watermark-based, trigger-based, and passport-based approaches, EncryIP demonstrates superior effectiveness in both training protected models and efficiently detecting the unauthorized spread of ML models.

NeurIPS Conference 2024 Conference Paper

Ensemble Learning for Heterogeneous Large Language Models with Deep Parallel Collaboration

• Yichong Huang
• Xiaocheng Feng
• Baohang Li
• Yang Xiang
• Hui Wang
• Ting Liu
• Bing Qin

Large language models (LLMs) exhibit complementary strengths in various tasks, motivating the research of LLM ensembling. However, existing work focuses on training an extra reward model or fusion model to select or combine all candidate answers, posing a great challenge to the generalization on unseen data distributions. Besides, prior methods use textual responses as communication media, ignoring the valuable information in the internal representations. In this work, we propose a training-free ensemble framework \textsc{DeePEn}, fusing the informative probability distributions yielded by different LLMs at each decoding step. Unfortunately, the vocabulary discrepancy between heterogeneous LLMs directly makes averaging the distributions unfeasible due to the token misalignment. To address this challenge, \textsc{DeePEn} maps the probability distribution of each model from its own probability space to a universal \textit{relative space} based on the relative representation theory, and performs aggregation. Next, we devise a search-based inverse transformation to transform the aggregated result back to the probability space of one of the ensembling LLMs (main model), in order to determine the next token. We conduct extensive experiments on ensembles of different number of LLMs, ensembles of LLMs with different architectures, and ensembles between the LLM and the specialist model. Experimental results show that (i) \textsc{DeePEn} achieves consistent improvements across six benchmarks covering subject examination, reasoning, and knowledge, (ii) a well-performing specialist model can benefit from a less effective LLM through distribution fusion, and (iii) \textsc{DeePEn} has complementary strengths with other ensemble methods such as voting.

IJCAI Conference 2024 Conference Paper

Meta-Learning via PAC-Bayesian with Data-Dependent Prior: Generalization Bounds from Local Entropy

• Shiyu Liu
• Wei Shi
• Zenglin Xu
• Shaogao Lv
• Yehong Zhang
• Hui Wang

Meta-learning accelerates the learning process on unseen learning tasks by acquiring prior knowledge through previous related tasks. The PAC-Bayesian theory provides a theoretical framework to analyze the generalization of meta-learning to unseen tasks. However, previous works still encounter two notable limitations: (1) they merely focus on the data-free priors, which often result in inappropriate regularization and loose generalization bounds; (2) more importantly, their optimization process usually involves nested optimization problems, incurring significant computational costs. To address these issues, we derive new generalization bounds and introduce a novel PAC-Bayesian framework for meta-learning that integrates data-dependent priors. This framework enables the extraction of optimal posteriors for each task in closed form, thereby allowing us to minimize generalization bounds incorporated data-dependent priors with only a simple local entropy. The resulting algorithm, which employs SGLD for sampling from the optimal posteriors, is stable, efficient, and computationally lightweight, eliminating the need for nested optimization. Extensive experimental results demonstrate that our proposed method outperforms the other baselines.

ECAI Conference 2024 Conference Paper

Model Provenance via Model DNA

• Xin Mu
• Yu Wang
• Yehong Zhang
• Jiaqi Zhang
• Hui Wang
• Yang Xiang
• Yue Yu

Understanding the life cycle of the machine learning (ML) model is an intriguing area of research (e. g. , understanding where the model comes from, how it is trained, and how it is used). Our focus is on a novel problem within this domain, namely Model Provenance (MP). MP concerns the relationship between a target model and its pre-training model and aims to determine whether a source model serves as the provenance for a target model. In this paper, we formulate this new challenge as a learning problem, supplementing our exploration with empirical discussions on its connections to existing works. Following that, we introduce “Model DNA”, an interesting concept encoding the model’s training data and input-output information to create a compact machine-learning model representation. Capitalizing on this model DNA, we establish an efficient framework consisting of three key components: DNA generation, DNA similarity loss, and a provenance classifier, aimed at identifying model provenance. We conduct evaluations on both computer vision and natural language processing tasks using various models, datasets, and scenarios to demonstrate the effectiveness of our approach.

YNICL Journal 2023 Journal Article

7T amygdala and hippocampus subfields in volumetry-based associations with memory: A 3-year follow-up study of early Alzheimer’s disease

• Laura Göschel
• Lea Kurz
• Andrea Dell'Orco
• Theresa Köbe
• Peter Körtvélyessy
• Ariane Fillmer
• Semiha Aydin
• Layla Tabea Riemann

INTRODUCTION: The hippocampus is the most prominent single region of interest (ROI) for the diagnosis and prediction of Alzheimer's disease (AD). However, its suitability in the earliest stages of cognitive decline, i.e., subjective cognitive decline (SCD), remains uncertain which warrants the pursuit of alternative or complementary regions. The amygdala might be a promising candidate, given its implication in memory as well as other psychiatric disorders, e.g. depression and anxiety, which are prevalent in SCD. In this 7 tesla (T) magnetic resonance imaging (MRI) study, we aimed to compare the contribution of volumetric measurements of the hippocampus, the amygdala, and their respective subfields, for early diagnosis and prediction in an AD-related study population. METHODS: Participants from a longitudinal study were grouped into SCD (n = 29), mild cognitive impairment (MCI, n = 23), AD (n = 22) and healthy control (HC, n = 31). All participants underwent 7T MRI at baseline and extensive neuropsychological testing at up to three visits (baseline n = 105, 1-year n = 78, 3-year n = 39). Analysis of covariance (ANCOVA) was used to assess group differences of baseline volumes of the amygdala and the hippocampus and their subfields. Linear mixed models were used to estimate the effects of baseline volumes on yearly changes of a z-scaled memory score. All models were adjusted to age, sex and education. RESULTS: Compared to the HC group, individuals with SCD showed smaller amygdala ROI volumes (range across subfields -11% to -1%), but not hippocampus ROI volumes (-2% to 1%) except for the hippocampus-amygdala-transition-area (-7%). However, cross-sectional associations between baseline memory and volumes were smaller for amygdala ROIs (std. ß [95% CI] ranging between 0.16 [0.08; 0.25] and 0.46 [0.31; 0.60]) than hippocampus ROIs (between 0.32 [0.19; 0.44] and 0.53 [0.40; 0.67]). Further, the association of baseline volumes with yearly memory change in the HC and SCD groups was similarly weak for amygdala ROIs and hippocampus ROIs. In the MCI group, volumes of amygdala ROIs were associated with a relevant yearly memory decline [95% CI] ranging between -0.12 [-0.24; 0.00] and -0.26 [-0.42; -0.09] for individuals with 20% smaller volumes than the HC group. However, effects were stronger for hippocampus ROIs with a corresponding yearly memory decline ranging between -0.21 [-0.35; -0.07] and -0.31 [-0.50; -0.13]. CONCLUSION: Volumes of amygdala ROIs, as determined by 7T MRI, might contribute to objectively and non-invasively identify patients with SCD, and thus aid early diagnosis and treatment of individuals at risk to develop dementia due to AD, however associations with other psychiatric disorders should be evaluated in further studies. The amygdala's value in the prediction of longitudinal memory changes in the SCD group remains questionable. Primarily in patients with MCI, memory decline over 3 years appears to be more strongly associated with volumes of hippocampus ROIs than amygdala ROIs.

NeurIPS Conference 2023 Conference Paper

Birder: Communication-Efficient 1-bit Adaptive Optimizer for Practical Distributed DNN Training

• Hanyang Peng
• Shuang Qin
• Yue Yu
• Jin Wang
• Hui Wang
• Ge Li

Various gradient compression algorithms have been proposed to alleviate the communication bottleneck in distributed learning, and they have demonstrated effectiveness in terms of high compression ratios and theoretical low communication complexity. However, when it comes to practically training modern deep neural networks (DNNs), these algorithms have yet to match the inference performance of uncompressed SGD-momentum (SGDM) and adaptive optimizers (e. g. ,Adam). More importantly, recent studies suggest that these algorithms actually offer no speed advantages over SGDM/Adam when used with common distributed DNN training frameworks ( e. g. , DistributedDataParallel (DDP)) in the typical settings, due to heavy compression/decompression computation or incompatibility with the efficient All-Reduce or the requirement of uncompressed warmup at the early stage. For these reasons, we propose a novel 1-bit adaptive optimizer, dubbed *Bi*nary *r*andomization a*d*aptive optimiz*er* (**Birder**). The quantization of Birder can be easily and lightly computed, and it does not require warmup with its uncompressed version in the beginning. Also, we devise Hierarchical-1-bit-All-Reduce to further lower the communication volume. We theoretically prove that it promises the same convergence rate as the Adam. Extensive experiments, conducted on 8 to 64 GPUs (1 to 8 nodes) using DDP, demonstrate that Birder achieves comparable inference performance to uncompressed SGDM/Adam, with up to ${2. 5 \times}$ speedup for training ResNet-50 and ${6. 3\times}$ speedup for training BERT-Base. Code is publicly available at https: //openi. pcl. ac. cn/c2net_optim/Birder.

AAMAS Conference 2023 Conference Paper

Cedric: A Collaborative DDoS Defense System Using Credit

• Jiawei Li
• Hui Wang
• Jilong Wang

Distributed denial of service (DDoS) is one of the most common and damaging cyber attacks, and its impact grows rapidly with the massive use of Internet. Collaborative DDoS defense across countries enables faster and more efficient DDoS attack mitigation. Collaboration requires countries that are not target victims to help detect and block the malicious flow, but selfish countries may refuse to do so because lacking individual gain compared with individual cost. In this paper, we model a stochastic game where selfish countries interact repeatedly and form coalitions to defend DDoS attacks. We design a multi-agent system, Cedric, to simulate and solve this complex stochastic game. Each agent adopts Q-learning to find their long-term optimal strategies, and credits are used to encourage efficient collaboration. The Shapley Value based reward assignment of Cedric satisfies several desired properties about fairness and stability. Simulations with trace data of over 7 years’ global DDoS attacks support the superiority of Cedric empirically.

EAAI Journal 2023 Journal Article

Diagnosis of hepatocellular carcinoma using deep network with multi-view enhanced patterns mined in contrast-enhanced ultrasound data

• Xiangfei Feng
• Wenjia Cai
• Rongqin Zheng
• Lina Tang
• Jianhua Zhou
• Hui Wang
• Jintang Liao
• Baoming Luo

Hepatocellular carcinoma, representing the most frequent primary liver cancer, is a common cancer disease that is the fourth leading cause of cancer-related mortality worldwide. In comparison, non-hepatocellular carcinoma liver cancers often present different prognoses and require distinct management which makes the accurate discrimination between hepatocellular carcinoma and non-hepatocellular carcinoma malignant lesions in contrast-enhanced ultrasound data critical for precise intervention. However, different types of liver cancers have similar enhanced patterns against the perfusion stages that raise the difficulty in the classification of hepatocellular carcinoma with the other liver cancers, especially when the contrast-enhanced ultrasound data is collected from different imaging machines. To this end, this paper innovatively proposes to extract perfusion features from a multi-view learning procedure for obtaining the inherent distinguishing features among liver cancers, leading to a more precise deep model in differentiating the hepatocellular carcinoma from other malignant cases. In particular, the proposed network consists of two novel structures for learning the correlation information among the different views to enhance the robustness of the features and fuse them by reducing redundant information. The proposed method is verified on a multi-source dataset collected from 1241 participants and achieves an AUC value of 89% for classification performance. The experimental results demonstrate the effectiveness of the proposed method for the diagnosis of hepatocellular carcinoma with a multi-source contrast-enhanced ultrasound dataset and might provide an effective assistant for clinical radiologists in liver cancer differentiation.

EAAI Journal 2023 Journal Article

Doctor selection based on aspect-based sentiment analysis and neutrosophic TOPSIS method

• Xihua Li
• Yun Luo
• Hui Wang
• Jiong Lin
• Bin Deng

Background: Physician-rating sites have become a convenient platform for patients to choose doctors. However, selecting an appropriate doctor through numerous online reviews is challengeable for patients. Although studies show that patients have different preferences for the aspects of healthcare services, existing doctor ranking methods rarely consider such preference information. Besides, they seldom handle the neutral sentiment information in patient reviews. Methods: To better assist patients in doctor selection, we propose a novel decision-making method that combines aspect-based sentiment analysis, single-valued neutrosophic sets and an extended Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) method. The method utilizes Word2Vec to construct a feature dictionary of patient satisfaction. A rule-based approach is employed to extract the aspects and the related sentiments from patients’ text reviews. Moreover, it takes advantage of single-valued neutrosophic sets to address the positive, neutral and negative sentiment information. On this basis, we rank the doctors with an extended TOPSIS method considering the patient’s attitudinal character and the preference information. Results: A case study on a review dataset demonstrates the stability and flexibility of the method. The comparative analysis reveals that our method can efficiently reduce information loss and it is more practical than existing methods. Conclusion: Considering patients’ different preferences for the aspects of healthcare services, the method can better provide healthcare decision support for patients.

JBHI Journal 2023 Journal Article

Dynamic Corrected Split Federated Learning With Homomorphic Encryption for U-Shaped Medical Image Networks

• Ziyuan Yang
• Yingyu Chen
• Huijie Huangfu
• Maosong Ran
• Hui Wang
• Xiaoxiao Li
• Yi Zhang

U-shaped networks have become prevalent in various medical image tasks such as segmentation, and restoration. However, most existing U-shaped networks rely on centralized learning which raises privacy concerns. To address these issues, federated learning (FL) and split learning (SL) have been proposed. However, achieving a balance between the local computational cost, model privacy, and parallel training remains a challenge. In this articler, we propose a novel hybrid learning paradigm called D ynamic Corrected S plit F ederated L earning ( DC-SFL ) for U-shaped medical image networks. To preserve data privacy, including the input, model parameters, label and output simultaneously, we propose to split the network into three parts hosted by different parties. We propose a D ynamic W eight C orrection S trategy ( DWCS ) to stabilize the training process and avoid the model drift problem due to data heterogeneity. To further enhance privacy protection and establish a trustworthy distributed learning paradigm, we propose to introduce additively homomorphic encryption into the aggregation process of client-side model, which helps prevent potential collusion between parties and provides a better privacy guarantee for our proposed method. The proposed DC-SFL is evaluated on various medical image tasks, and the experimental results demonstrate its effectiveness. In comparison with state-of-the-art distributed learning methods, our method achieves competitive performance.

EAAI Journal 2023 Journal Article

Dynamic model-driven intelligent fault diagnosis method for rotary vector reducers

• Junkang Zheng
• Hui Wang
• Anil Kumar
• Jiawei Xiang

The diagnosis of faults in rotary vector (RV) reducers using machine data-driven artificial intelligence (AI) models plays an important role, but it is difficult to obtain complete fault sample labeled data. Without labeled data, AI-based intelligent fault diagnosis models will fail. To solve the problem of data scarcity, a lumped parameter model of an RV reducer is developed to produce a sufficient training sample for AI models. First, a lumped parameter model of the healthy RV reducer is constructed and updated by the Pearson correlation coefficient (PCC) technique to obtain an agreeable dynamic model with a certain precision. Then, mathematical expressions of numerous fault modes with different fault severities are inserted into the model to calculate the fault samples. The simulated failure samples serve as training samples of AI-based intelligent models. Finally, CNN, VGG and ResNet are selected as the representatives of AI model, and then unknown fault samples are identified by applying data from real-time machinery. The experimental results suggest that the present method can be used to overcome the problem of insufficient fault samples in RV reducers.

EAAI Journal 2023 Journal Article

Extended belief rule base with ensemble imbalanced learning for lymph node metastasis diagnosis in endometrial carcinoma

• Long-Hao Yang
• Tian-Yu Ren
• Fei-Fei Ye
• Haibo Hu
• Hui Wang
• Hui Zheng

Lymph node metastasis (LNM) constitutes one of the main prognostic factors for long-term survival in endometrial carcinoma (EC). However, the previous studies on LNM diagnosis failed to consider both model interpretability and class imbalance. In this study, the extended belief rule base (EBRB) expert system is introduced to develop a novel EBRB-based LNM diagnosis model. First, the interpretability of the EBRB expert system is investigated to demonstrate the feasibility on LNM diagnosis; Second, imbalanced learning is introduced to improve rule generation scheme for constructing base EBRBs; Third, by considering the trust of base EBRBs and base diagnoses, ensemble learning is introduced to improve rule inference scheme for diagnosing final LNM. In the case study, real EC patient data collected from Fujian Provincial Maternity and Children's Hospital are used to verify the effectiveness of the proposed EBRB-based model by comparing with the variants of rule generation schemes and rule inference schemes, as well as some machine learning-based LNM diagnosis models. The comparative results showed that the proposed EBRB-based model has better sensitivity, specificity, and geometric mean in diagnosing LNM for EC patients.

JMLR Journal 2023 Journal Article

FedLab: A Flexible Federated Learning Framework

• Dun Zeng
• Siqi Liang
• Xiangjing Hu
• Hui Wang
• Zenglin Xu

FedLab is a lightweight open-source framework for the simulation of federated learning. The design of FedLab focuses on federated learning algorithm effectiveness and communication efficiency. It allows customization on server optimization, client optimization, communication agreement, and communication compression. Also, FedLab is scalable in different deployment scenarios with different computation and communication resources. We hope FedLab could provide flexible APIs as well as reliable baseline implementations and relieve the burden of implementing novel approaches for researchers in the FL community. The source code, tutorial, and documentation can be found at https://github.com/SMILELab-FL/FedLab. [abs] [ pdf ][ bib ] [ code ] &copy JMLR 2023. ( edit, beta )

JBHI Journal 2023 Journal Article

Multimodal Data Matters: Language Model Pre-Training Over Structured and Unstructured Electronic Health Records

• Sicen Liu
• Xiaolong Wang
• Yongshuai Hou
• Ge Li
• Hui Wang
• Hui Xu
• Yang Xiang
• Buzhou Tang

As two important textual modalities in electronic health records (EHR), both structured data (clinical codes) and unstructured data (clinical narratives) have recently been increasingly applied to the healthcare domain. Most existing EHR-oriented studies, however, either focus on a particular modality or integrate data from different modalities in a straightforward manner, which usually treats structured and unstructured data as two independent sources of information about patient admission and ignore the intrinsic interactions between them. In fact, the two modalities are documented during the same encounter where structured data inform the documentation of unstructured data and vice versa. In this paper, we proposed a Medical Multimodal Pre-trained Language Model, named MedM-PLM, to learn enhanced EHR representations over structured and unstructured data and explore the interaction of two modalities. In MedM-PLM, two Transformer-based neural network components are firstly adopted to learn representative characteristics from each modality. A cross-modal module is then introduced to model their interactions. We pre-trained MedM-PLM on the MIMIC-III dataset and verified the effectiveness of the model on three downstream clinical tasks, i. e. , medication recommendation, 30-day readmission prediction and ICD coding. Extensive experiments demonstrate the power of MedM-PLM compared with state-of-the-art methods. Further analyses and visualizations show the robustness of our model, which could potentially provide more comprehensive interpretations for clinical decision-making.

JBHI Journal 2023 Journal Article

SHAPE: A Sample-Adaptive Hierarchical Prediction Network for Medication Recommendation

• Sicen Liu
• Xiaolong Wang
• Jingcheng Du
• Yongshuai Hou
• Xianbing Zhao
• Hui Xu
• Hui Wang
• Yang Xiang

Effectively medication recommendation with complex multimorbidity conditions is a critical yet challenging task in healthcare. Most existing works predicted medications based on longitudinal records, which assumed the encoding format of intra-visit medical events are serialized and information transmitted patterns of learning longitudinal sequence data are stable. However, the following conditions may have been ignored: 1) A more compact encoder for intra-relationship in the intra-visit medical event is urgent; 2) Strategies for learning accurate representations of the variable longitudinal sequences of patients are different. In this article, we proposed a novel Sample-adaptive Hierarchical medicAtion Prediction nEtwork, termed SHAPE, to tackle the above challenges in the medication recommendation task. Specifically, we design a compact intra-visit set encoder to encode the relationship in the medical event for obtaining visit-level representation and then develop an inter-visit longitudinal encoder to learn the patient-level longitudinal representation efficiently. To endow the model with the capability of modeling the variable visit length, we introduce a soft curriculum learning method to assign the difficulty of each sample automatically by the visit length. Extensive experiments on a benchmark dataset verify the superiority of our model compared with several state-of-the-art baselines.

YNIMG Journal 2022 Journal Article

A novel algorithm for multiplicative speckle noise reduction in ex vivo human brain OCT images

• Divya Varadarajan
• Caroline Magnain
• Morgan Fogarty
• David A. Boas
• Bruce Fischl
• Hui Wang

Optical coherence tomography (OCT) images of ex vivo human brain tissue are corrupted by multiplicative speckle noise that degrades the contrast to noise ratio (CNR) of microstructural compartments. This work proposes a novel algorithm to reduce noise corruption in OCT images that minimizes the penalized negative log likelihood of gamma distributed speckle noise. The proposed method is formulated as a majorize-minimize problem that reduces to solving an iterative regularized least squares optimization. We demonstrate the usefulness of the proposed method by removing speckle in simulated data, phantom data and real OCT images of human brain tissue. We compare the proposed method with state of the art filtering and non-local means based denoising methods. We demonstrate that our approach removes speckle accurately, improves CNR between different tissue types and better preserves small features and edges in human brain tissue.

YNICL Journal 2022 Journal Article

Brain microstructural antecedents of visual difficulties in infants born very preterm

• Rahul Chandwani
• Karen Harpster
• Julia E. Kline
• Ved Mehta
• Hui Wang
• Stephanie L. Merhar
• Terry L. Schwartz
• Nehal A. Parikh

Infants born very preterm (VPT) are at risk of later visual problems. Although neonatal screening can identify ophthalmologic abnormalities, subtle perinatal brain injury and/or delayed brain maturation may be significant contributors to complex visual-behavioral problems. Our aim was to assess the micro and macrostructural antecedents of early visual-behavioral difficulties in VPT infants by using diffusion MRI (dMRI) at term-equivalent age. We prospectively recruited a cohort of 262 VPT infants (≤32 weeks gestational age [GA]) from five neonatal intensive care units. We obtained structural and diffusion MRI at term-equivalent age and administered the Preverbal Visual Assessment (PreViAs) questionnaire to parents at 3-4 months corrected age. We used constrained spherical deconvolution to reconstruct nine white matter tracts of the visual pathways with high reliability and performed fixel-based analysis to derive fiber density (FD), fiber-bundle cross-section (FC), and combined fiber density and cross-section (FDC). In multiple logistic regression analyses, we related these tract metrics to visual-behavioral function. Of 262 infants, 191 had both high-quality dMRI and completed PreViAs, constituting the final cohort: mean (SD) GA was 29.3 (2.4) weeks, 90 (47.1%) were males, and postmenstrual age (PMA) at MRI was 42.8 (1.3) weeks. FD and FC of several tracts were altered in infants with (N = 59) versus those without retinopathy of prematurity (N = 132). FDC of the left posterior thalamic radiations (PTR), left inferior longitudinal fasciculus (ILF), right superior longitudinal fasciculus (SLF), and left inferior fronto-occipital fasciculus (IFOF) were significantly associated with visual attention scores, prior to adjusting for confounders. After adjustment for PMA at MRI, GA, severe retinopathy of prematurity, and total brain volume, FDC of the left PTR, left ILF, and left IFOF remained significantly associated with visual attention. Early visual-behavioral difficulties in VPT infants are preceded by micro and macrostructural abnormalities in several major visual pathways at term-equivalent age.

AIIM Journal 2022 Journal Article

CATNet: Cross-event attention-based time-aware network for medical event prediction

• Sicen Liu
• Xiaolong Wang
• Yang Xiang
• Hui Xu
• Hui Wang
• Buzhou Tang

Medical event prediction (MEP) is a fundamental task in the healthcare domain, which needs to predict medical events, including medications, diagnosis codes, laboratory tests, procedures, outcomes, and so on, according to historical medical records of patients. Many researchers have tried to build MEP models to overcome the challenges caused by the heterogeneous and irregular temporal characteristics of EHR data. However, most of them consider the heterogenous and temporal medical events separately and ignore the correlations among different types of medical events, especially relations between heterogeneous historical medical events and target medical events. In this paper, we propose a novel neural network based on attention mechanism called Cross-event Attention-based Time-aware Network (CATNet) for MEP. It is a time-aware, event-aware and task-adaptive method with the following advantages: 1) modeling heterogeneous information and temporal information in a unified way and considering irregular temporal characteristics locally and globally respectively, 2) taking full advantage of correlations among different types of events via cross-event attention. Experiments on two public datasets (MIMIC-III and eICU) show CATNet outperforms other state-of-the-art methods on various MEP tasks. The source code of CATNet is released at https: //github. com/sherry6247/CATNet. git.

NeurIPS Conference 2022 Conference Paper

Exploring evolution-aware & -free protein language models as protein function predictors

• Mingyang Hu
• Fajie Yuan
• Kevin Yang
• Fusong Ju
• Jin Su
• Hui Wang
• Fei Yang
• Qiuyang Ding

Large-scale Protein Language Models (PLMs) have improved performance in protein prediction tasks, ranging from 3D structure prediction to various function predictions. In particular, AlphaFold, a ground-breaking AI system, could potentially reshape structural biology. However, the utility of the PLM module in AlphaFold, Evoformer, has not been explored beyond structure prediction. In this paper, we investigate the representation ability of three popular PLMs: ESM-1b (single sequence), MSA-Transformer (multiple sequence alignment), and Evoformer (structural), with a special focus on Evoformer. Specifically, we aim to answer the following key questions: (1) Does the Evoformer trained as part of AlphaFold produce representations amenable to predicting protein function? (2) If yes, can Evoformer replace ESM-1b and MSA-Transformer? (3) How much do these PLMs rely on evolution-related protein data? In this regard, are they complementary to each other? We compare these models by empirical study along with new insights and conclusions. All code and datasets for reproducibility are available at https: //github. com/elttaes/Revisiting-PLMs.

IS Journal 2021 Journal Article

Anomalous Event Sequence Detection

• Boxiang Dong
• Zhengzhang Chen
• Lu-An Tang
• Haifeng Chen
• Hui Wang
• Kai Zhang
• Ying Lin
• Zhichun Li

Anomaly detection has been widely applied in modern data-driven security applications to detect abnormal events/entities that deviate from the majority. However, less work has been done in terms of detecting suspicious event sequences/paths, which are better discriminators than single events/entities for distinguishing normal and abnormal behaviors in complex systems such as cyber-physical systems. A key and challenging step in this endeavor is how to discover those abnormal event sequences from millions of system event records in an efficient and accurate way. To address this issue, we propose NINA, a network diffusion based algorithm for identifying anomalous event sequences. Experimental results on both static and streaming data show that NINA is efficient (processes about 2 million records per minute) and accurate.

YNIMG Journal 2021 Journal Article

High-fidelity approximation of grid- and shell-based sampling schemes from undersampled DSI using compressed sensing: Post mortem validation

• Robert Jones
• Chiara Maffei
• Jean Augustinack
• Bruce Fischl
• Hui Wang
• Berkin Bilgic
• Anastasia Yendiki

While many useful microstructural indices, as well as orientation distribution functions, can be obtained from multi-shell dMRI data, there is growing interest in exploring the richer set of microstructural features that can be extracted from the full ensemble average propagator (EAP). The EAP can be readily computed from diffusion spectrum imaging (DSI) data, at the cost of a very lengthy acquisition. Compressed sensing (CS) has been used to make DSI more practical by reducing its acquisition time. CS applied to DSI (CS-DSI) attempts to reconstruct the EAP from significantly undersampled q-space data. We present a post mortem validation study where we evaluate the ability of CS-DSI to approximate not only fully sampled DSI but also multi-shell acquisitions with high fidelity. Human brain samples are imaged with high-resolution DSI at 9.4T and with polarization-sensitive optical coherence tomography (PSOCT). The latter provides direct measurements of axonal orientations at microscopic resolutions, allowing us to evaluate the mesoscopic orientation estimates obtained from diffusion MRI, in terms of their angular error and the presence of spurious peaks. We test two fast, dictionary-based, L2-regularized algorithms for CS-DSI reconstruction. We find that, for a CS acceleration factor of R=3, i.e., an acquisition with 171 gradient directions, one of these methods is able to achieve both low angular error and low number of spurious peaks. With a scan length similar to that of high angular resolution multi-shell acquisition schemes, this CS-DSI approach is able to approximate both fully sampled DSI and multi-shell data with high accuracy. Thus it is suitable for orientation reconstruction and microstructural modeling techniques that require either grid- or shell-based acquisitions. We find that the signal-to-noise ratio (SNR) of the training data used to construct the dictionary can have an impact on the accuracy of CS-DSI, but that there is substantial robustness to loss of SNR in the test data. Finally, we show that, as the CS acceleration factor increases beyond R=3, the accuracy of these reconstruction methods degrade, either in terms of the angular error, or in terms of the number of spurious peaks. Our results provide useful benchmarks for the future development of even more efficient q-space acceleration techniques.

YNIMG Journal 2021 Journal Article

Quantification of volumetric morphometry and optical property in the cortex of human cerebellum at micrometer resolution

• Chao J. Liu
• William Ammon
• Viviana Siless
• Morgan Fogarty
• Ruopeng Wang
• Alessia Atzeni
• Iman Aganj
• Juan Eugenio Iglesias

The surface of the human cerebellar cortex is much more tightly folded than the cerebral cortex. Volumetric analysis of cerebellar morphometry in magnetic resonance imaging studies suffers from insufficient resolution, and therefore has had limited impact on disease assessment. Automatic serial polarization-sensitive optical coherence tomography (as-PSOCT) is an emerging technique that offers the advantages of microscopic resolution and volumetric reconstruction of large-scale samples. In this study, we reconstructed multiple cubic centimeters of ex vivo human cerebellum tissue using as-PSOCT. The morphometric and optical properties of the cerebellar cortex across five subjects were quantified. While the molecular and granular layers exhibited similar mean thickness in the five subjects, the thickness varied greatly in the granular layer within subjects. Layer-specific optical property remained homogenous within individual subjects but showed higher cross-subject variability than layer thickness. High-resolution volumetric morphometry and optical property maps of human cerebellar cortex revealed by as-PSOCT have great potential to advance our understanding of cerebellar function and diseases.

YNIMG Journal 2020 Journal Article

Insight into the fundamental trade-offs of diffusion MRI from polarization-sensitive optical coherence tomography in ex vivo human brain

• Robert Jones
• Giorgia Grisot
• Jean Augustinack
• Caroline Magnain
• David A. Boas
• Bruce Fischl
• Hui Wang
• Anastasia Yendiki

In the first study comparing high angular resolution diffusion MRI (dMRI) in the human brain to axonal orientation measurements from polarization-sensitive optical coherence tomography (PSOCT), we compare the accuracy of orientation estimates from various dMRI sampling schemes and reconstruction methods. We find that, if the reconstruction approach is chosen carefully, single-shell dMRI data can yield the same accuracy as multi-shell data, and only moderately lower accuracy than a full Cartesian-grid sampling scheme. Our results suggest that current dMRI reconstruction approaches do not benefit substantially from ultra-high b-values or from very large numbers of diffusion-encoding directions. We also show that accuracy remains stable across dMRI voxel sizes of 1 ​mm or smaller but degrades at 2 ​mm, particularly in areas of complex white-matter architecture. We also show that, as the spatial resolution is reduced, axonal configurations in a dMRI voxel can no longer be modeled as a small set of distinct axon populations, violating an assumption that is sometimes made by dMRI reconstruction techniques. Our findings have implications for in vivo studies and illustrate the value of PSOCT as a source of ground-truth measurements of white-matter organization that does not suffer from the distortions typical of histological techniques.

IJCAI Conference 2019 Conference Paper

Dual Visual Attention Network for Visual Dialog

• Dan Guo
• Hui Wang
• Meng Wang

Visual dialog is a challenging task, which involves multi-round semantic transformations between vision and language. This paper aims to address cross-modal semantic correlation for visual dialog. Motivated by that Vg (global vision), Vl (local vision), Q (question) and H (history) have inseparable relevances, the paper proposes a novel Dual Visual Attention Network (DVAN) to realize (Vg, Vl, Q, H)--> A. DVAN is a three-stage query-adaptive attention model. In order to acquire accurate A (answer), it first explores the textual attention, which imposes the question on history to pick out related context H'. Then, based on Q and H', it implements respective visual attentions to discover related global image visual hints Vg' and local object-based visual hints Vl'. Next, a dual crossing visual attention is proposed. Vg' and Vl' are mutually embedded to learn the complementary of visual semantics. Finally, the attended textual and visual features are combined to infer the answer. Experimental results on the VisDial v0. 9 and v1. 0 datasets validate the effectiveness of the proposed approach.

TCS Journal 2019 Journal Article

Quantifying sequential subsumption

• Hui Wang
• Cees H. Elzinga
• Zhiwei Lin
• Jordan Vincent

Subsumption is used in knowledge representation and ontology to describe the relationship between concepts. Concept A is subsumed by concept B if the extension of A is always a subset of the extension of B, irrespective of the interpretation. The subsumption relation is also useful in other data analysis tasks such as pattern recognition – for example in image analysis to detect objects in an image, and in spectral data analysis to detect the presence of a reference pattern in a given spectrum. Sometimes the subsumption relation may not be 100% true, so it is useful to quantify this relationship. In this paper we study how to quantify subsumption for sequential patterns. We review existing work on subsumption, give an axiomatic characterisation of subsumption, and present one general approach to quantification in terms of set intersection operation over concept extension. Constructing the concept extension set explicitly is impossible without specifying the domain of discourse and the interpretation. Instead, we focus on concept intension for sequences as patterns and propose to represent concept intension of a sequence by its subsequences. We further consider different types of concept intension set – subsequence set, subsequence multiset, embedding set and embedding set with constraints such as warping and selection. We then present a general algorithmic framework for computing set intersections, and specific algorithms for computing different concept intension sets. We also present an experimental evaluation of these algorithms with regard to their runtime performance.

AIIM Journal 2019 Journal Article

Segmentation of breast MR images using a generalised 2D mathematical model with inflation and deflation forces of active contours

• Andrik Rampun
• Bryan W. Scotney
• Philip J. Morrow
• Hui Wang
• John Winder

In medical computer aided diagnosis systems, image segmentation is one of the major pre-processing steps used to ensure only the region of interest, such as the breast region, will be processed in subsequent steps. Nevertheless, breast segmentation is a difficult task due to low contrast and inhomogeneity, especially when estimating the chest wall in magnetic resonance (MR) images. In fact, the chest wall comprises fat, skin, muscles, and the thoracic skeleton, which can misguide automatic methods when attempting to estimate its location. The objective of the study is to develop a fully automated method for breast and pectoral muscle boundary estimation in MR images. Firstly, we develop a 2D breast mathematical model based on 30 MRI slices (from a patient) and identify important landmarks to obtain a model for the general shape of the breast in an axial plane. Subsequently, we use Otsu's thresholding approach and Canny edge detection to estimate the breast boundary. The active contour method is then employed using both inflation and deflation forces to estimate the pectoral muscle boundary by taking account of information obtained from the proposed 2D model. Finally, the estimated boundary is smoothed using a median filter to remove outliers. Our two datasets contain 60 patients in total and the proposed method is evaluated based on 59 patients (one patient is used to develop the 2D breast model). On the first dataset (9 patients) the proposed method achieved Jaccard = 81. 1% ±6. 1 % and dice coefficient= 89. 4% ±4. 1 % and on the second dataset (50 patients) Jaccard = 84. 9% ±5. 8 % and dice coefficient = 92. 3% ±3. 6 %. These results are qualitatively comparable with the existing methods in the literature.

YNIMG Journal 2018 Journal Article

as-PSOCT: Volumetric microscopic imaging of human brain architecture and connectivity

• Hui Wang
• Caroline Magnain
• Ruopeng Wang
• Jay Dubb
• Ani Varjabedian
• Lee S. Tirrell
• Allison Stevens
• Jean C. Augustinack

Polarization sensitive optical coherence tomography (PSOCT) with serial sectioning has enabled the investigation of 3D structures in mouse and human brain tissue samples. By using intrinsic optical properties of back-scattering and birefringence, PSOCT reliably images cytoarchitecture, myeloarchitecture and fiber orientations. In this study, we developed a fully automatic serial sectioning polarization sensitive optical coherence tomography (as-PSOCT) system to enable volumetric reconstruction of human brain samples with unprecedented sample size and resolution. The 3. 5 μm in-plane resolution and 50 μm through-plane voxel size allow inspection of cortical layers that are a single-cell in width, as well as small crossing fibers. We show the abilities of as-PSOCT in quantifying layer thicknesses of the cerebellar cortex and creating microscopic tractography of intricate fiber networks in the subcortical nuclei and internal capsule regions, all based on volumetric reconstructions. as-PSOCT provides a viable tool for studying quantitative cytoarchitecture and myeloarchitecture and mapping connectivity with microscopic resolution in the human brain.

JBHI Journal 2018 Journal Article

New Precision Metrics for Contrast Sensitivity Testing

• Michael Dorr
• Tobias Elze
• Hui Wang
• Zhong-Lin Lu
• Peter J. Bex
• Luis A. Lesmes

Visual sensitivity is comprehensively described by the contrast sensitivity function (CSF), but current routine clinical care does not include its assessment because of the time-consuming need to estimate thresholds for a large number of spatial frequencies. The quick CSF method, however, dramatically reduces testing times by using a Bayesian information maximization rule. We evaluate the test-retest variability of a tablet-based quick CSF implementation in a study with 100 subjects who repeatedly assessed their vision with and without optical correction. We first discuss two commonly used measures of repeatability, intraclass correlation and the Bland-Altman Coefficient of Repeatability, and show that they are vulnerable to artifacts. Instead, we propose to formulate precision as an information retrieval task: from all repeat test scores, can we retrieve a certain individual based on their first test score? We then use rank-based analyses such as mean average precision as a better measure to compare different test metrics, and show that the highest test-retest precision is achieved using a summary statistic, the area under the log CSF (AULCSF). This demonstrates the benefit of assessment of the whole CSF compared to sensitivity at individual spatial frequencies only. AULCSF also yields best discrimination performance (99. 2%) between measurements that were taken with and without glasses, respectively, even better than CSF Acuity. The tablet-based quick CSF thus enables the rapid and reliable home monitoring of visual function, which has the potential to improve early diagnosis and treatment of ophthalmic pathologies such as diabetic retinopathy or age-related macular degeneration.

IJCAI Conference 2018 Conference Paper

Progressive Blockwise Knowledge Distillation for Neural Network Acceleration

• Hui Wang
• Hanbin Zhao
• Xi Li
• Xu Tan

As an important and challenging problem in machine learning and computer vision, neural network acceleration essentially aims to enhance the computational efficiency without sacrificing the model accuracy too much. In this paper, we propose a progressive blockwise learning scheme for teacher-student model distillation at the subnetwork block level. The proposed scheme is able to distill the knowledge of the entire teacher network by locally extracting the knowledge of each block in terms of progressive blockwise function approximation. Furthermore, we propose a structure design criterion for the student subnetwork block, which is able to effectively preserve the original receptive field from the teacher network. Experimental results demonstrate the effectiveness of the proposed scheme against the state-of-the-art approaches.

IS Journal 2014 Journal Article

Collaboration Pattern and Topic Analysis on Intelligence and Security Informatics Research

• Wenli Liu
• Xiaolong Zheng
• Tao Wang
• Hui Wang

In this article, researcher collaboration patterns and research topics on Intelligence and Security Informatics (ISI) are investigated using social network analysis approaches. The collaboration networks exhibit scale-free property and small-world effect. From these networks, the authors obtain the key researchers, institutions, and three important topics.

YNIMG Journal 2014 Journal Article

Cross-validation of serial optical coherence scanning and diffusion tensor imaging: A study on neural fiber maps in human medulla oblongata

• Hui Wang
• Junfeng Zhu
• Martin Reuter
• Louis N. Vinke
• Anastasia Yendiki
• David A. Boas
• Bruce Fischl
• Taner Akkin

We established a strategy to perform cross-validation of serial optical coherence scanner imaging (SOCS) and diffusion tensor imaging (DTI) on a postmortem human medulla. Following DTI, the sample was serially scanned by SOCS, which integrates a vibratome slicer and a multi-contrast optical coherence tomography rig for large-scale three-dimensional imaging at microscopic resolution. The DTI dataset was registered to the SOCS space. An average correlation coefficient of 0. 9 was found between the co-registered fiber maps constructed by fractional anisotropy and retardance contrasts. Pixelwise comparison of fiber orientations demonstrated good agreement between the DTI and SOCS measures. Details of the comparison were studied in regions exhibiting a variety of fiber organizations. DTI estimated the preferential orientation of small fiber tracts; however, it didn't capture their complex patterns as SOCS did. In terms of resolution and imaging depth, SOCS and DTI complement each other, and open new avenues for cross-modality investigations of the brain.

YNIMG Journal 2014 Journal Article

Serial optical coherence scanner for large-scale brain imaging at microscopic resolution

• Hui Wang
• Junfeng Zhu
• Taner Akkin

We describe a serial optical coherence scanner (SOCS) for high resolution imaging of ex-vivo brain. SOCS integrates a multi-contrast optical coherence tomography and a vibratome slicer to establish comprehensive brain anatomy and fiber pathways in three-dimensional space. Rat brain images are demonstrated by utilizing intrinsic optical contrasts including back-scattering, birefringence and optic axis orientation, which are simultaneously generated from the same dataset. Volumetric images from serial scans are combined to realize large scale brain maps. Nerve fiber tracts are globally described in 3D by retardance, and delicately delineated by cross-polarization at the resolution of 15×15×5. 5μm3. In-plane orientations of the tracts are quantified by optic axis orientation. SOCS offers a new solution for complete reconstructions of macroscopic tissues such as primate and human brains at microscopic resolution. The technique also opens up varieties of opportunities for connectome studies and systematic investigations on neurological diseases and brain disorders.

TCS Journal 2013 Journal Article

Versatile string kernels

• Cees H. Elzinga
• Hui Wang

This paper proposes a class of string kernels that can handle a variety of subsequence-based features. Slight adaptations of the basic algorithm allow for weighing subsequence lengths, restricting or soft-penalizing gap-size, character-weighing and soft-matching of characters. An easy extension of the kernels allows for comparing run-length encoded strings with a time-complexity that is independent of the length of the original strings. Such kernels have applications in image processing, computational biology, in demography and in comparing partial rankings.

YNIMG Journal 2012 Journal Article

A tale of two itches. Common features and notable differences in brain activation evoked by cowhage and histamine induced itch

• Alexandru D.P. Papoiu
• Robert C. Coghill
• Robert A. Kraft
• Hui Wang
• Gil Yosipovitch

Previous PET and fMRI brain imaging studies targeting neural networks processing itch sensation have used histamine as the sole itch inducer. In contrast with histamine, cowhage-induced itch is mediated via proteinase activated receptors PAR2 and is transmitted through a separate spinothalamic pathway, therefore imaging the brain activation evoked by cowhage could provide further insight into central processing of itch. We report for the first time a functional MRI Arterial Spin Labeling (ASL) study of neuronal processing of itch induced by cowhage, analyzed in contrast with histamine-induced itch. We also explored the brain responses induced by histamine and cowhage combined in a tight sequence. The results of our analyses obtained in a group of 15 healthy volunteers suggested that cowhage and histamine co-activated a core group of brain structures, while also revealing notable differences. Core areas activated by both stimuli were found in the thalamus, primary and secondary somatosensory cortices, posterior parietal cortex, superior and middle temporal cortices, PCC, ACC, precuneus and cuneus. Cowhage induced a notably distinct and more extensive involvement of the insular cortex, claustrum, basal ganglia, putamen, thalamic nuclei and pulvinar. The differences observed between these two itch modalities were investigated to determine the impact of quantitative versus qualitative factors, and correlations between itch intensity and the patterns in brain activation were explored. Our analysis revealed that the most significant differences between cowhage and histamine itch were not affected by stimulus intensity, although a subset of regions displayed activations which were intensity-dependent. The combined application of cowhage and histamine highlighted the role of insula and claustrum in the processing of both itch modalities in the same time. The present results suggest the existence of overlapping but also distinct neuronal networks processing these two different types of itch.

YNIMG Journal 2011 Journal Article

Reconstructing micrometer-scale fiber pathways in the brain: Multi-contrast optical coherence tomography based tractography

• Hui Wang
• Adam J. Black
• Junfeng Zhu
• Tyler W. Stigen
• Muhammad K. Al-Qaisi
• Theoden I. Netoff
• Aviva Abosch
• Taner Akkin

Comprehensive understanding of connective neural pathways in the brain has put great challenges on the current imaging techniques, for which three-dimensional (3D) visualization of fiber tracts with high spatiotemporal resolution is desirable. Here we present optical imaging and tractography of rat brain ex-vivo using multi-contrast optical coherence tomography (MC-OCT), which is capable of simultaneously generating depth-resolved images of reflectivity, phase retardance, optic axis orientation and, for in-vivo studies, blood flow images. Using the birefringence property of myelin sheath, nerve fiber tracts as small as a few tens of micrometers can be resolved and neighboring fiber tracts with different orientations can be distinguished in cross-sectional optical slices, 2D en-face images and 3D volumetric images. Combinational contrast of MC-OCT images enables visualization of the spatial architecture and nerve fiber orientations in the brain with unprecedented detail. The results suggest that optical tractography, by virtue of its direct accessibility to nerve fibers, has the potential to validate diffusion magnetic resonance images and investigate structural connections in normal brain and neurological disorders. In addition, an endoscopic MC-OCT may be useful in neurosurgical interventions to aid in placement of deep brain stimulating electrodes.

TCS Journal 2008 Journal Article

Algorithms for subsequence combinatorics

• Cees Elzinga
• Sven Rahmann
• Hui Wang

A subsequence is obtained from a string by deleting any number of characters; thus in contrast to a substring, a subsequence is not necessarily a contiguous part of the string. Counting subsequences under various constraints has become relevant to biological sequence analysis, to machine learning, to coding theory, to the analysis of categorical time series in the social sciences, and to the theory of word complexity. We present theorems that lead to efficient dynamic programming algorithms to count (1) distinct subsequences in a string, (2) distinct common subsequences of two strings, (3) matching joint embeddings in two strings, (4) distinct subsequences with a given minimum span, and (5) sequences generated by a string allowing characters to come in runs of a length that is bounded from above.

IJCAI Conference 2007 Conference Paper

• Hui Wang

Time series data abounds in real world problems. Measuring the similarity of time series is a key to solving these problems. One state of the art measure is the longest common subsequence. This measure advocates using the length of the longest common subsequence as an indication of similarity between sequences, but ignores information contained in the second, third, .. ., longest subsequences. In order to capture the common information in sequences maximally we propose a novel measure of sequence similarity -- the number of all common subsequences. We show that this measure satisfies the common properties of similarity functions. Calculating this measure is not trivial as a brute force approach is exponential in time. We present a novel dynamic programming algorithm to calculate this number in polynomial time. We also suggest a different way of extending a class of such measures to multidimensional, real-valued time series, in the spirit of probabilistic metric spaces. We conducted an experimental study on the new similarity measure and the extension method for classification. It was found that both the new similarity and the extension method are consistently competitive.

IJCAI Conference 2005 Conference Paper

A flexible and robust similarity measure based on contextual probability

• Hui Wang
• Werner

Arguably, analogy is one of the most important aspects of intelligent reasoning. It has been hypothesized that, given suitable background knowledge, analogy can be viewed as a logical inference process. This study follows another school of thought that argues that similarity can provide a probabilistic basis for inference and analogy. Most similarity measures (which are frequently viewed as being conceptually equivalent to distance measures) are restricted to either nominal or ordinal attributes, and some are confined to classification tasks. This paper proposes a flexible similarity measure that is task-independent and applies to both nominal and ordinal data in a conceptually uniform way. The proposed similarity measure is derived from a probability function and corresponds to the intuition that if we consider all neighborhoods around a data point, the data points closer to this point should be included in more of these neighborhoods than more distant points. Experiments we have conducted to demonstrate the usefulness of this measure indicate that it fares very competitively with commonly used similarity measures.

TCS Journal 2001 Journal Article

A relation – algebraic approach to the region connection calculus

• Ivo Düntsch
• Hui Wang
• Steve McCloskey

We explore the relation – algebraic aspects of the region connection calculus (RCC) of Randell et al. , Proceedings of the CADE, vol 11, pp. 786–790, Springer, Berlin, 1992a. In particular, we present a refinement of the RCC8 table which shows that the axioms provide for more relations than are listed in the present table. We also show that each RCC model leads to a Boolean algebra. Finally, we prove that a refined version of the RCC5 table has as models all atomless Boolean algebras B with the natural ordering as the “part-of” relation, and that the table is closed under first-order definable relations iff B is homogeneous.

IROS Conference 2001 Conference Paper

Embedding cooperation in robots to play soccer game

• Hui Wang
• Han Wang 0001
• Chunmiao Wang
• William Y. C. Soh

Robotic soccer provides an opportunity to explore such a challenging research topic that multiple agents (physical robots or sofbots) work together in a realtime, noisy and adversarial environment to obtain specific objectives. It requires each agent can not only deal with infinite unpredictable situations, but also present cooperation with others. The previous researches about cooperation often put emphasis on task decomposition and conflict avoidance among team members. In this paper, we describe a robot architecture, which addresses "scaling cooperation" among robots, and meanwhile keeps each robot making decision independently. The architecture is based on "ideal cooperation" principle and implemented for Small Robot League in RoboCup Experimental results prove its effectiveness and reveal several primary characteristics of behaviors in robotic soccer. Finally, some important problems of future work are discussed.

IJCAI Conference 1999 Conference Paper

A Lattice Machine Approach to Automated Casebase Design: Marrying Lazy and Eager Learning

• Hui Wang
• Werner Dubitzky
• Ivo Duntsch
• David Bell

Case-based reasoning (CBR) is concerned with solving new problems by adapting solutions that worked for similar problems in the past. Years of experience in building and fielding C B R systems have shown that the "rase approach" is not free from problems. It has been realized that the knowledge engineering effort required for designing many real-world easebases can be prohibitively high. Based on the wide-spread use of databases and powerful machine learning methods, some C B R researchers have been investigating the possibility of designing casebases automatically. This paper proposes a flexible model for the automatic discovery of abstract cases from data. bases based on the Lattice Machine. It also proposes an efficient and effective algorithm for retrieving such cases. Besides the known benefits associated with abstract cases, the main advantages of this approach are that the discovery process is fully automated (no knowledge engineering costs). K e y w o r d s: case-based reasoning, machine learning, knowledge acquisition, automated modeling

TCS Journal 1994 Journal Article

Some results concerning 2-D on-line tessellation acceptors and 2-D alternating finite automata

• Tao Jiang
• Oscar H. Ibarra
• Hui Wang

A two-dimensional nondeterministic on-line tessellation acceptor (2-NOTA) is a special type of real-time two-dimensional nondeterministic cellular automaton in which data flows from the upper-left corner to the lower-right corner. A two-dimensional alternating finite automaton (2-AFA) is an alternating finite automaton with a two-dimensional rectangular input whose input head can move in all four directions on the input. In this paper, we show that 2-NOTAs and 2-AFAs are incomparable. This answers in the negative an open question posed by Ito (this journal, 1989). Closure properties of the classes of languages (i. e. , sets of two-dimensional patterns) accepted by two-way, three-way, and four-way two-dimensional alternating finite automata and two-dimensional alternating finite automata with only universal states (2-UFAs) are also obtained which answer several open questions posed by Inoue and Takanami (1988).

TCS Journal 1992 Journal Article

A characterization of exponential-time languages by alternating context-free grammars

• Oscar H. Ibarra
• Tao Jiang
• Hui Wang

We show that the class of exponential-time languages or, equivalently, the class of languages accepted by alternating pushdown automata (APDAs), is exactly the class of languages generated by linear-erasing alternating context-free grammars (ACFGs). An ACFG is generalization of an ordinary context-free grammar in which we allow the use of universal nonterminals in much the same way as universal states are used in APDAs. It was recently claimed in [9] that APDAs are equivalent to ACFGs. However, the proofs in both directions have major flaws which do not seem to be correctable. As it turns out, the proof of the claim does not follow from a simple extension of the well-known constructions for the nonalternating case. Our proof is, in fact, for a modified claim: APDAs are equivalent to linear-erasing ACFGs, where linear-erasing means that there is a constant c such that every string of length n in the language generated by the ACFG has a derivation in which all intermediate sentential forms are at most cn long.

TCS Journal 1991 Journal Article

Parallel parsing on a one-way linear array of finite-state machines

• Oscar H. Ibarra
• Tao Jiang
• Hui Wang

Efficient parallel algorithms for some parsing problems are presented. These problems include the parsing of linear context-free languages, languages accepted by nondeterministic one-counter automata, and transductions defined by a special class of two-tape nondeterministic finite-state transducers. The model of parallel computation is a one-way linear array of identical finite-state machines. The data movement in the array is one-way, from left to right. For inputs of length n, the array uses n nodes. Our algorithms can actually produce a parse, i. e. a sequence of rules (moves) that generates (accepts) an input, in linear time. When only a no/yes answer is required, the parsing problem becomes a recognition problem. The best serial (RAM) algorithms for the corresponding recognition problems take O(n 2/log2 n) time and space. Previous parallel algorithms for the recognition problems run in linear time on a one-way linear array of finite-state machines.