Jian Zhou

EAAI Journal 2026 Journal Article

A novel attention-based long short-term memory latency prediction model for stream processing applications

• Zheng Chu
• Dongwen Chen
• Xinfeng Zhang
• Baozhu Li
• Jiong Yu
• Xusheng Du
• Jian Zhou
• Weiyun Li

The proliferation of the Internet of Things has led to a significant increase in the number and types of smart devices, resulting in an exponential growth of streaming data volume and types. Consequently, many fields have adopted stream processing application (SPA) to handle real-time processing scenarios. Practitioners and scientists across various domains rely on latency prediction of these applications, which is essential for performance analysis and proactive optimization. However, predicting the latency of SPAs remains challenging due to their diverse types and internal complexity. To accurately predict the latency, this paper explores related work extensively and investigates the directed acyclic graph characteristics of such applications. Three kinds of features (i. e. , application features, data features, and system features) are identified and summarized as major factors influencing latency. Furthermore, we propose and implement a artificial intelligence-based real-time prediction framework in Apache Flink, a popular stream processing system, for application latency prediction. This framework collects three types of real-time metrics, constructs features, and utilizes an attention-based long short-term memory recurrent neural network model to accurately predict latency at run-time. Experimental results from six benchmarks show that the proposed model accurately predicts latency using the identified features. More importantly, our model outperforms the state-of-the-art model in terms of prediction error and accuracy due to the proposed real-time framework and attention mechanism. Furthermore, our model can achieve high prediction accuracy for newly developed applications in a short time.

EAAI Journal 2026 Journal Article

Corrigendum to “A novel attention-based long short-term memory latency prediction model for stream processing applications” [Eng. Appl. Artif. Intell. 163 (2026) 112890]

• Zheng Chu
• Dongwen Chen
• Xinfeng Zhang
• Baozhu Li
• Jiong Yu
• Xusheng Du
• Jian Zhou
• Weiyun Li

EAAI Journal 2026 Journal Article

Deep learning-aided Laser Doppler Velocimeter-Inertial Measurement Unit Fusion for Robust Vehicle Localization in Global Navigation Satellite Systems-denied environments

• Zhiyi Xiang
• Qi Wang
• Xiaoming Nie
• Jian Zhou

Achieving reliable and precise vehicle positioning is paramount for modern autonomous systems, yet it remains a formidable challenge in Global Navigation Satellite Systems (GNSS)-denied environments, especially when relying on ubiquitous low-cost Micro-Electro-Mechanical Systems (MEMS) Inertial Measurement Units (IMUs). This paper introduces a solution that enhances MEMS IMU capabilities by integrating two symmetrically mounted dual-beam Laser Doppler Velocimeters (LDVs). Our core innovation lies in leveraging two specialized Long Short-Term Memory (LSTM) networks that robustly regress the vehicle’s yaw and lateral velocities by effectively fusing both LDV and IMU outputs. To further elevate system accuracy, we propose an LDV outlier handling strategy and a method for LSTM prediction reliability detection designed to mitigate the adverse effects of anomalous network outputs. The vehicle velocities from the LDVs, augmented by our LSTM-derived yaw and lateral velocities, are then fused with MEMS IMU data within a Lie group-based Kalman filter. Experimental validation through two rigorous test sets demonstrates that our method significantly reduces system positioning errors under prolonged GNSS-denied conditions, outperforming existing LDV-based methods. This work underscores the potential of combining precise LDV measurements with the predictive power of deep learning and a robust Lie group-based data fusion strategy for accurate and reliable autonomous vehicle localization.

EAAI Journal 2026 Journal Article

Machine learning to enhance strain-resilience humidity sensing on flexible surface acoustic wave platform

• Yanhong Xia
• Zhangbin Ji
• Jian Zhou
• Yihao Guo
• Hui Chen
• Jinbo Zhang
• Yongqing Fu

Flexible surface acoustic wave (SAW) humidity sensors have garnered considerable attention in fields such as environmental monitoring and healthcare, mainly attributed to their advantages such as wearability, applicability in non-planar scenarios, quasi-digital output, and wireless passive capabilities. However, improvement in performance of these flexible SAW humidity sensors faces great challenges such as low electromechanical coupling coefficient, poor humidity response or sensitivity, and introduction of detection errors caused by mechanical strain interference. Herein, we developed a flexible SAW humidity sensor utilizing an aluminum scandium nitride (AlScN) piezoelectric film deposited on ultrathin glass substrates, incorporating ternary nanocomposites of graphene quantum dots-polyethyleneimine-silica nanoparticles (GQDs-PEI-SiO2 NPs) as the sensitive layers, which demonstrated an ultra-high sensitivity of 5. 02 kHz (kHz)/%Relative Humidity (RH). To address critical issues of strain interferences under randomly bending or deformation conditions, we applied machine learning (ML) algorithms to establish correlations between sensor's response signal features and humidity labels, thereby effectively mitigating unreliable humidity measurements caused by significant strain interferences, with improved precision and specificity. After comprehensive evaluation and analysis using various artificial intelligence algorithms, multilayer perceptron regression model was identified as the best performer in humidity prediction under strain interferences, with a coefficient of determination as high as 0. 997 and a mean square error of ∼0. 479. Reliability and generalization capabilities of this model were verified, and such the strategy not only significantly enhances the performance metrics of flexible humidity sensors but also provides an innovative and precision solution under various strain interferences using the flexible SAW sensors.

EAAI Journal 2025 Journal Article

A novel hybrid data-driven domain generalization approach with dual-perspective feature fusion for intelligent fault diagnosis

• Lanjun Wan
• Jian Zhou
• Jiaen Ning
• Yuanyuan Li
• Changyun Li

Domain generalization-based fault diagnosis (DGFD) approaches do not require access to the target domain during model training, but they usually rely on numerous labeled source domain data. However, only few labeled source domain data can be obtained in actual diagnosis scenarios. Therefore, a novel hybrid data-driven domain generalization (DG) approach with dual-perspective feature fusion for intelligent fault diagnosis (FD) is proposed. Firstly, to solve the problem of scarce training samples in the source domains, the rolling bearing (RB) and the gear simulated vibration models are established to generate numerous labeled simulated vibration data, and the improved auxiliary classifier generative adversarial network (ACGAN) is used to effectively balance the simulated and real data. Secondly, a simulated and real data-driven DG network that fuses intra-domain invariant features and mutually-invariant features between domains (SRDGN-IM) is proposed, where the intra-domain invariant features are learned through distillation idea and the mutually-invariant features are learned through adversarial training, which can make the diagnosis model better learn the key generalization features from source domains to obtain more accurate diagnosis results. Finally, a series of DG experiments are conducted on the gearbox and bearing datasets, and the average FD accuracies of the proposed approach reach 87. 45% and 89. 10% respectively under different DG tasks.

AAAI Conference 2025 Conference Paper

BSDB-Net: Band-Split Dual-Branch Network with Selective State Spaces Mechanism for Monaural Speech Enhancement

• Cunhang Fan
• Enrui Liu
• Andong Li
• Jianhua Tao
• Jian Zhou
• Jiahao Li
• Chengshi Zheng
• Zhao Lv

Although the complex spectrum-based speech enhancement (SE) methods have achieved significant performance, coupling amplitude and phase can lead to a compensation effect, where amplitude information is sacrificed to compensate for the phase that is harmful to SE. In addition, to further improve the performance of SE, many modules are stacked onto SE, resulting in increased model complexity that limits the application of SE. To address these problems, we proposed a dual-path network based on compressed frequency using Mamba. First, we extract amplitude and phase information through parallel dual branches. This approach leverages structured complex spectra to implicitly capture phase information and solves the compensation effect by decoupling amplitude and phase, and the network incorporates an interaction module to suppress unnecessary parts and recover missing components from the other branch. Second, to reduce network complexity, the network introduces a band-split strategy to compress the frequency dimension. To further reduce complexity while maintaining good performance, we designed a Mamba-based module that models the time and frequency dimensions under linear complexity. Finally, compared to baselines, our model achieves an average 8.3 times reduction in computational complexity while maintaining superior performance. Furthermore, it achieves a 25 times reduction in complexity compared to transformer-based models.

ICML Conference 2025 Conference Paper

Inverse Flow and Consistency Models

• Yuchen Zhang
• Jian Zhou

Inverse generation problems, such as denoising without ground truth observations, is a critical challenge in many scientific inquiries and real-world applications. While recent advances in generative models like diffusion models, conditional flow matching, and consistency models achieved impressive results by casting generation as denoising problems, they cannot be directly used for inverse generation without access to clean data. Here we introduce Inverse Flow (IF), a novel framework that enables using these generative models for inverse generation problems including denoising without ground truth. Inverse Flow can be flexibly applied to nearly any continuous noise distribution and allows complex dependencies. We propose two algorithms for learning Inverse Flows, Inverse Flow Matching (IFM) and Inverse Consistency Model (ICM). Notably, to derive the computationally efficient, simulation-free inverse consistency model objective, we generalized consistency training to any forward diffusion processes or conditional flows, which have applications beyond denoising. We demonstrate the effectiveness of IF on synthetic and real datasets, outperforming prior approaches while enabling noise distributions that previous methods cannot support. Finally, we showcase applications of our techniques to fluorescence microscopy and single-cell genomics data, highlighting IF’s utility in scientific problems. Overall, this work expands the applications of powerful generative models to inversion generation problems.

IJCAI Conference 2025 Conference Paper

ListenNet: A Lightweight Spatio-Temporal Enhancement Nested Network for Auditory Attention Detection

• Cunhang Fan
• Xiaoke Yang
• Hongyu Zhang
• Ying Chen
• Lu Li
• Jian Zhou
• Zhao Lv

Auditory attention detection (AAD) aims to identify the direction of the attended speaker in multi-speaker environments from brain signals, such as Electroencephalography (EEG) signals. However, existing EEG-based AAD methods overlook the spatio-temporal dependencies of EEG signals, limiting their decoding and generalization abilities. To address these issues, this paper proposes a Lightweight Spatio-Temporal Enhancement Nested Network (ListenNet) for AAD. The ListenNet has three key components: Spatio-temporal Dependency Encoder (STDE), Multi-scale Temporal Enhancement (MSTE), and Cross-Nested Attention (CNA). The STDE reconstructs dependencies between consecutive time windows across channels, improving the robustness of dynamic pattern extraction. The MSTE captures temporal features at multiple scales to represent both fine-grained and long-range temporal patterns. In addition, the CNA integrates hierarchical features more effectively through novel dynamic attention mechanisms to capture deep spatio-temporal correlations. Experimental results on three public datasets demonstrate the superiority of ListenNet over state-of-the-art methods in both subject-dependent and challenging subject-independent settings, while reducing the trainable parameter count by approximately 7 times. Code is available at: https: //github. com/fchest/ListenNet.

IJCAI Conference 2025 Conference Paper

M3ANet: Multi-scale and Multi-Modal Alignment Network for Brain-Assisted Target Speaker Extraction

• Cunhang Fan
• Ying Chen
• Jian Zhou
• Zexu Pan
• Jingjing Zhang
• Youdian Gao
• Xiaoke Yang
• Zhengqi Wen

The brain-assisted target speaker extraction (TSE) aims to extract the attended speech from mixed speech by utilizing the brain neural activities, for example Electroencephalography (EEG). However, existing models overlook the issue of temporal misalignment between speech and EEG modalities, which hampers TSE performance. In addition, the speech encoder in current models typically uses basic temporal operations (e. g. , one-dimensional convolution), which are unable to effectively extract target speaker information. To address these issues, this paper proposes a multi-scale and multi-modal alignment network (M3ANet) for brain-assisted TSE. Specifically, to eliminate the temporal inconsistency between EEG and speech modalities, the modal alignment module that uses a contrastive learning strategy is applied to align the temporal features of both modalities. Additionally, to fully extract speech information, multi-scale convolutions with GroupMamba modules are used as the speech encoder, which scans speech features at each scale from different directions, enabling the model to capture deep sequence information. Experimental results on three publicly available datasets show that the proposed model outperforms current state-of-the-art methods across various evaluation metrics, highlighting the effectiveness of our proposed method. The source code is available at: https: //github. com/fchest/M3ANet.

IJCAI Conference 2025 Conference Paper

MHANet: Multi-scale Hybrid Attention Network for Auditory Attention Detection

• Lu Li
• Cunhang Fan
• Hongyu Zhang
• Jingjing Zhang
• Xiaoke Yang
• Jian Zhou
• Zhao Lv

Auditory attention detection (AAD) aims to detect the target speaker in a multi-talker environment from brain signals, such as electroencephalography (EEG), which has made great progress. However, most AAD methods solely utilize attention mechanisms sequentially and overlook valuable multi-scale contextual information within EEG signals, limiting their ability to capture long-short range spatiotemporal dependencies simultaneously. To address these issues, this paper proposes a multi-scale hybrid attention network (MHANet) for AAD, which consists of the multi-scale hybrid attention (MHA) module and the spatiotemporal convolution (STC) module. Specifically, MHA combines channel attention and multi-scale temporal and global attention mechanisms. This effectively extracts multi-scale temporal patterns within EEG signals and captures long-short range spatiotemporal dependencies simultaneously. To further improve the performance of AAD, STC utilizes temporal and spatial convolutions to aggregate expressive spatiotemporal representations. Experimental results show that the proposed MHANet achieves state-of-the-art performance with fewer trainable parameters across three datasets, 3 times lower than that of the most advanced model. Code is available at: https: //github. com/fchest/MHANet.

IROS Conference 2025 Conference Paper

SparseMeXt: Unlocking the Potential of Sparse Representations for HD Map Construction

• Anqing Jiang
• Jinhao Chai
• Yu Gao 0042
• Yiru Wang 0001
• Yuwen Heng
• Zhigang Sun 0001
• Hao Sun
• Zezhong Zhao

Recent advancements in high-definition (HD) map construction have demonstrated the effectiveness of dense representations, which heavily rely on computationally intensive bird’s-eye view (BEV) features. While sparse representations offer a more efficient alternative by avoiding dense BEV processing, existing methods often lag behind due to the lack of tailored designs. These limitations have hindered the competitiveness of sparse representations in online HD map construction. In this work, we systematically revisit and enhance sparse representation techniques, identifying key architectural and algorithmic improvements that bridge the gap with—and ultimately surpass—dense approaches. We introduce a dedicated network architecture optimized for sparse map feature extraction, a sparse-dense segmentation auxiliary task to better leverage geometric and semantic cues, and a denoising module guided by physical priors to refine predictions. Through these enhancements, our method achieves state-of-the-art performance on the nuScenes dataset, significantly advancing HD map construction and centerline detection. Specifically, SparseMeXt-Tiny reaches a mean average precision (mAP) of 55. 5% at 32 frames per second (fps), while SparseMeXt-Base attains 65. 2% mAP. Scaling the backbone and decoder further, SparseMeXt-Large achieves an mAP of 68. 9% at over 20 fps, establishing a new benchmark for sparse representations in HD map construction. These results underscore the untapped potential of sparse methods, challenging the conventional reliance on dense representations and redefining efficiency-performance trade-offs in the field.

EAAI Journal 2024 Journal Article

Masked autoencoder with dynamic multi-loss adaptation mechanism for few shot wafer map pattern recognition

• Qi Liang
• Jian Zhou
• Yonglin Wang

Wafer Map Pattern Recognition (WMPR) is a critical aspect of semiconductor manufacturing. It indicates how to improve the manufacturing yields as we probe into the failure issues of the processes. In literature works, researchers often use balanced datasets with ample datapoints to address WMPR tasks, however, novel defects often emerge with few previous observations in real-world manufacturing. Unfortunately, efforts to solve WMPR problems in few-shot scenarios remain scanty. To bridge this gap, we define a new task, Few Shot Wafer Map Pattern Recognition(FSWMPR), which attempts to learning a classifier to distinguish unseen classes with only a few labeled instances available. In such a task, expeditiously learning transferable feature embeddings is extremely challenging. In this paper, we propose an innovative two-stage strategy to wrestle with the problem of FSWMPR. In the first stage, we leverage a masked autoencoder to obtain efficacious representations of defect wafer map images through reconstructing pixel values of masked patches based on smooth-l1 loss. In the second stage, we create a novel finetuning mechanism, “Dynamic Multi-Loss Adaptation Mechanism”, which utilize three cooperative losses to accelerate fast feature transfer for few-shot scenarios. Surprisingly, if three losses are reduced to one comparative loss, we still achieve more competitive accuracy than those meta-learning or finetuning methods, which is worth noting that our two stages involve no label information at all. Extensive experiments and analyses are conducted on WM811K datasets. Compared with other algorithms, our methods offer fresh solutions by creatively integrating self-supervised masked autoencoder with a novel finetune mechanism which is efficacious for FSWMPR.

ICML Conference 2023 Conference Paper

Dirichlet Diffusion Score Model for Biological Sequence Generation

• Pavel Avdeyev
• Chenlai Shi
• Yuhao Tan
• Kseniia Dudnyk
• Jian Zhou

Designing biological sequences is an important challenge that requires satisfying complex constraints and thus is a natural problem to address with deep generative modeling. Diffusion generative models have achieved considerable success in many applications. Score-based generative stochastic differential equations (SDE) model is a continuous-time diffusion model framework that enjoys many benefits, but the originally proposed SDEs are not naturally designed for modeling discrete data. To develop generative SDE models for discrete data such as biological sequences, here we introduce a diffusion process defined in the probability simplex space with stationary distribution being the Dirichlet distribution. This makes diffusion in continuous space natural for modeling discrete data. We refer to this approach as Dirchlet diffusion score model. We demonstrate that this technique can generate samples that satisfy hard constraints using a Sudoku generation task. This generative model can also solve Sudoku, including hard puzzles, without additional training. Finally, we applied this approach to develop the first human promoter DNA sequence design model and showed that designed sequences share similar properties with natural promoter sequences.

YNICL Journal 2022 Journal Article

Shared Transdiagnostic Neuroanatomical Signatures Across First-episode Patients with Major Psychiatric Diseases and Individuals at Familial Risk

• Linna Jia
• Xiaowei Jiang
• Qikun Sun
• Jian Zhou
• Linzi Liu
• Ting Sun
• Pengshuo Wang
• Yanqing Tang

BACKGROUND: Nowadays, increasing evidence has found transdiagnostic neuroimaging biomarkers across major psychiatric disorders (MPDs). However, it remains to be known whether this transdiagnostic pattern of abnormalities could also be seen in individuals at familial high-risk for MPDs (FHR). We aimed to examine shared neuroanatomical endophenotypes and protective biomarkers for MPDs. METHODS: This study examined brain grey matter volume (GMV) of individuals by voxel-based morphometry method. A total of 287 individuals were included, involving 100 first-episode medication-naive MPDs, 87 FHR, and 110 healthy controls (HC). They all underwent high-resolution structural magnetic resonance imaging (MRI). RESULTS: At the group level, we found MPDs were characterized by decreased GMV in the right fusiform gyrus, the right inferior occipital gyrus, and the left anterior and middle cingulate gyri compared to HC and FHR. Of note, the GMV of the left superior temporal gyrus was increased in FHR relative to MPDs and HC. At the subgroup level, the comparisons within the FHR group did not return any significant difference, and we found GMV difference among subgroups within the MPDs group only in the opercular part of the right inferior frontal gyrus. CONCLUSION: Together, our findings uncover common structural disturbances across MPDs and substantial changes in grey matter that may relate to high hereditary risk across FHR, potentially underscoring the importance of a transdiagnostic way to explore the neurobiological mechanisms of major psychiatric disorders.

YNICL Journal 2022 Journal Article

The correspondence between morphometric MRI and metabolic profile in Rasmussen’s encephalitis

• Chongyang Tang
• Peng Ren
• Kaiqiang Ma
• Siyang Li
• Xiongfei Wang
• Yuguang Guan
• Jian Zhou
• Tianfu Li

Volumetric magnetic resonance imaging (MRI) atrophy is a hallmark of Rasmussen's encephalitis (RE). Here, we aim to investigate voxel-wise gray matter (GM) atrophy in RE, and its associations with glucose hypometabolism and neurotransmitter distribution utilizing MRI and PET data. In this study, fifteen RE patients and fourteen MRI normal subjects were included in this study. Voxel-wise GM volume and glucose metabolic uptake were evaluated using structural MRI and FDG-PET images, respectively. Spatial Spearman's correlation was performed between GM atrophy of RE with FDG uptake alterations, and neurotransmitter distributions provided in the JuSpace toolbox. Compared with the control group, RE patients displayed extensive GM volume loss not only in the ipsilateral hemisphere, but also in the frontal lobe, basal ganglia, and cerebellum in the contralateral hemisphere. Within the RE group, the insular and temporal cortices exhibited significantly more GM atrophy on the ipsilesional than the contralesional side. FDG-PET data revealed significant hypometabolism in areas surrounding the insular cortices in the ipsilesional hemisphere. RE-related GM volumetric atrophy was spatially correlated with hypomebolism in FDG uptake, and with spatial distribution of the dopaminergic and serotonergic neurotransmitter systems. The spatial concordance of morphological changes with metabolic abnormalities suggest FDG-PET offers potential value for RE diagnosis. The GM alterations associated with neurotransmitter distribution map could provide novel insight in understanding the neuropathological mechanisms and clinical feature of RE.

IJCAI Conference 2021 Conference Paper

Novelty Detection via Contrastive Learning with Negative Data Augmentation

• Chengwei Chen
• Yuan Xie
• Shaohui Lin
• Ruizhi Qiao
• Jian Zhou
• Xin Tan
• Yi Zhang
• Lizhuang Ma

Novelty detection is the process of determining whether a query example differs from the learned training distribution. Previous generative adversarial networks based methods and self-supervised approaches suffer from instability training, mode dropping, and low discriminative ability. We overcome such problems by introducing a novel decoder-encoder framework. Firstly, a generative network (decoder) learns the representation by mapping the initialized latent vector to an image. In particular, this vector is initialized by considering the entire distribution of training data to avoid the problem of mode-dropping. Secondly, a contrastive network (encoder) aims to ``learn to compare'' through mutual information estimation, which directly helps the generative network to obtain a more discriminative representation by using a negative data augmentation strategy. Extensive experiments show that our model has significant superiority over cutting-edge novelty detectors and achieves new state-of-the-art results on various novelty detection benchmarks, e. g. CIFAR10 and DCASE. Moreover, our model is more stable for training in a non-adversarial manner, compared to other adversarial based novelty detection methods.

EAAI Journal 2021 Journal Article

Optimization of support vector machine through the use of metaheuristic algorithms in forecasting TBM advance rate

• Jian Zhou
• Yingui Qiu
• Shuangli Zhu
• Danial Jahed Armaghani
• Chuanqi Li
• Hoang Nguyen
• Saffet Yagiz

The advance rate (AR) of a tunnel boring machine (TBM) in hard rock condition is a key parameter for the successful accomplishment of a tunneling project, and the proper and reliable prediction of this parameter can lead to minimizing the risks associated to high capital costs and scheduling for such projects. This research aims at optimizing the hyper-parameters of the support vector machine (SVM) technique through the use of three optimization algorithms, namely, gray wolf optimization (GWO), whale optimization algorithm (WOA) and moth flame optimization (MFO), in forecasting TBM AR. In fact, the role of these optimization techniques is to optimize the hyperparameters ‘C’ and ‘gamma’ of the SVM model to get higher performance prediction. To develop the hybrid SVM-based models, 1, 286 sample sets of data collected from a water transfer tunnel in Malaysia comprising seven input variables, i. e. , rock mass rating, uniaxial compressive strength, Brazilian tensile strength, rock quality designation, weathering zone, thrust force and revolution per minute, and one output variable, i. e. , TBM AR, were considered and used. Several GWO-SVM, WOA-SVM and MFO-SVM models were constructed to predict TBM AR considering their effective parameters. The accuracy levels of the proposed models were assessed using four statistical indices, i. e. , the coefficient of determination (R2), root mean squared error (RMSE), mean absolute error (MAE), and variance accounted for (VAF). Modeling results revealed that the MFO algorithm can capture better hyper-parameters of the SVM model in predicting TBM AR among all three hybrid models. R2 of (0. 9623 and 0. 9724), RMSE of (0. 1269 and 0. 1155), and VAF of (96. 24 and 97. 34%), respectively, for training and test stages of the MFO-SVM model confirmed that this hybrid SVM model is a powerful and applicable technique addressing problems related to TBM performance with a high level of accuracy.

AAAI Conference 2021 Conference Paper

Temporal Segmentation of Fine-gained Semantic Action: A Motion-Centered Figure Skating Dataset

• Shenglan Liu
• Aibin Zhang
• Yunheng Li
• Jian Zhou
• Li Xu
• Zhuben Dong
• Renhao Zhang

Temporal Action Segmentation (TAS) has achieved great success in many fields such as exercise rehabilitation, movie editing, etc. Currently, task-driven TAS is a central topic in human action analysis. However, motion-centered TAS, as an important topic, is little researched due to unavailable datasets. In order to explore more models and practical applications of motion-centered TAS, we introduce a Motion-Centered Figure Skating (MCFS) dataset in this paper. Compared with existing temporal action segmentation datasets, the MCFS dataset is fine-grained semantic, specialized and motion-centered. Besides, RGB-based and Skeletonbased features are provided in the MCFS dataset. Experimental results show that existing state-of-the-art methods are difficult to achieve excellent segmentation results (including accuracy, edit and F1 score) in the MCFS dataset. This indicates that MCFS is a challenging dataset for motioncentered TAS. The latest dataset can be downloaded at https: //shenglanliu. github. io/mcfs-dataset/.

ECAI Conference 2020 Conference Paper

Joint 3D Face Reconstruction and Dense Face Alignment via Deep Face Feature Alignment

• Jian Zhou
• Zhangjin Huang

Reconstructing a 3D face from a single face image is a challenging problem in a wide range of applications. Due to the lack of a large number of 3D face datasets with ground truth, previous methods usually adopt weakly supervised learning methods. However, most methods only utilize pixel level information, which causes the convolutional neural network models to easily fall into local minima. This paper proposes a novel method of 3D face reconstruction and dense face alignment based on a single face image under unknown pose, expression and illumination. We not only consider the difference between the input face image and the rendered image at the pixel level, but also consider their difference in the deep feature space. First, a 3D face model is constructed from a single face image by using a parameterized face model. Then, the 3D face model is rendered to a 2D plane through a differentiable renderer. Next, the correspondences between the input face image and the rendered image in the pixel space and the deep feature space are established, respectively. Finally, our model is trained by back propagation. Experiments on AFLW2000-3D and AFLW-LFPA show that the proposed method outperforms existing approaches in both 3D face reconstruction and dense face alignment.

YNICL Journal 2017 Journal Article

Identification of the epileptogenic zone of temporal lobe epilepsy from stereo-electroencephalography signals: A phase transfer entropy and graph theory approach

• Meng-yang Wang
• Jing Wang
• Jian Zhou
• Yu-guang Guan
• Feng Zhai
• Chang-qing Liu
• Fei-fei Xu
• Yi-xian Han

The aim of this research is to apply an approach based on phase transfer entropy (PTE) and graph theory to study the interactions between the stereo-electroencephalography (SEEG) activities recorded in multilobar origin, in order to evaluate their ability to detect the epileptogenic zone (EZ) of temporal lobe epilepsies (TLE). Forty-three patients were included in this retrospective study. Five to sixteen (median = 12) multilead electrodes were implanted per patient, and, for each patient, a sub-set of between 10 and 32 (median = 22) bipolar derivations was selected for analysis. The leads were classified into the onset leads (OLs), the early propagation leads (EPLs), and the rest of the leads (RLs). The results showed that a significantly different dynamic trend of the out/in ratio (more obvious in the gamma band) distinguishes the OLs from RLs in the 23 patients who were seizure-free not only during the ictal event (significant elevation), but also during the inter-,pre-, late-ictal periods, and especially in the post-ictal (sharp decline) state. However, in the 20 patients who were not-seizure-free, the differences between the OLs and RLs during the post-ictal period were not found in any frequency band. The dynamic trend was used to predict surgical outcome, and the results showed that the sensitivity was 91% and the specificity was 70%. In brief, this study indicates that our approach may add new and valuable information, providing efficient quantitative measures useful for localizing the EZ.

EAAI Journal 2016 Journal Article

Optimizing h value for fuzzy linear regression with asymmetric triangular fuzzy coefficients

• Fangning Chen
• Yizeng Chen
• Jian Zhou
• Yuanyuan Liu

The parameter h in a fuzzy linear regression model is vital since it influences the degree of the fitting of the estimated fuzzy linear relationship to the given data directly. However, it is usually subjectively pre-selected by a decision-maker as an input to the model in practice. In Liu and Chen (2013), a new concept of system credibility was introduced by combining the system fuzziness with the system membership degree, and a systematic approach was proposed to optimize the h value for fuzzy linear regression analysis using the minimum fuzziness criterion with symmetric triangular fuzzy coefficients. As an extension, in this paper, their approach is extended to asymmetric cases, and the procedure to find the optimal h value to maximize the system credibility of the fuzzy linear regression model with asymmetric triangular fuzzy coefficients is described. Some illustrative examples are given to show the detailed procedure of this approach, and comparative studies are also conducted via the testing data sets.

EAAI Journal 2015 Journal Article

Fuzzy linear regression models for QFD using optimized h values

• Yuanyuan Liu
• Yizeng Chen
• Jian Zhou
• Shuya Zhong

In recent years, the fuzzy linear regression (FLR) approach is widely applied in the quality function deployment (QFD) to identify the vague and inexact functional relationships between the customer requirements and the engineering characteristics on account of its advantages of objectiveness and reality. However, the h value, which is a vital parameter in the proceeding of the FLR model, is usually set by the design team subjectively. In this paper, we propose a systematic approach using the FLR models attached with optimized h values to identify the functional relationships in QFD, where the coefficients are assumed as symmetric triangular fuzzy numbers. The h values in the FLR models are determined according to the criterion of maximizing the system credibilities of the FLR models. Furthermore, an illustrative example is provided to demonstrate the performance of the proposed approach. Results of the numerical example show that the fuzzy coefficients obtained through the FLR models with optimized h values are more effective than those obtained through the FLR models with arbitrary h values selected by the design team.

ICML Conference 2014 Conference Paper

Deep Supervised and Convolutional Generative Stochastic Network for Protein Secondary Structure Prediction

• Jian Zhou
• Olga G. Troyanskaya

Predicting protein secondary structure is a fundamental problem in protein structure prediction. Here we present a new supervised generative stochastic network (GSN) based method to predict local secondary structure with deep hierarchical representations. GSN is a recently proposed deep learning technique (Bengio & Thibodeau-Laufer, 2013) to globally train deep generative model. We present the supervised extension of GSN, which learns a Markov chain to sample from a conditional distribution, and applied it to protein structure prediction. To scale the model to full-sized, high-dimensional data, like protein sequences with hundreds of amino-acids, we introduce a convolutional architecture, which allows efficient learning across multiple layers of hierarchical representations. Our architecture uniquely focuses on predicting structured low-level labels informed with both low and high-level representations learned by the model. In our application this corresponds to labeling the secondary structure state of each amino-acid residue. We trained and tested the model on separate sets of non-homologous proteins sharing less than 30% sequence identity. Our model achieves 66. 4% Q8 accuracy on the CB513 dataset, better than the previously reported best performance 64. 9% (Wang et al. , 2011) for this challenging secondary structure prediction problem.