Author name cluster

Juan Liu

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

14 papers

2 author rows

AAAI Conference 2026 Conference Paper

Decoupled Spatiotemporal Forecasting from Extreme Sparse Observations via Quantized Latent Space

Zhongnan Weng
Yue Hong
Hang Yu
Jiayi Que
Juan Liu
Xiangrong Liu

Predicting spatiotemporal fields governed by partial differential equations (PDEs) from sparse sensor data is a critical and long-standing challenge in science and engineering. Recent deep learning approaches, particularly neural operators, have shown considerable promise in solving PDEs. However, their performance degrades significantly in the demanding regime of extreme sparsity, characterized by spatial sensor coverage of less than 1% and limited temporal observations. To overcome this limitation, we propose a novel framework that decouples the task into two stages: spatial reconstruction and temporal extrapolation. In the first stage, rather than reconstructing the high-dimensional physical field directly, our model learns to reconstruct the complete latent features from sparse observations—features that would otherwise be extracted from a dense field. This process is stabilized by a Vector Quantization (VQ) bottleneck, which discretizes the latent space. In the second stage, a decoder-only Transformer performs temporal extrapolation by autoregressively predicting the future sequence of these discrete latent indices. This design inherently allows the model to generalize to new initial conditions and varying forecast horizons, akin to standard autoregressive models. We validate our framework on three challenging benchmarks, achieving state-of-the-art (SOTA) performance under severe sparsity constraints. Furthermore, we introduce a challenging benchmark dataset based on fire dynamics simulations. On this benchmark, our model successfully forecasts the field's evolution 30 frames into the future from a single timeframe with less than 0.1% spatial observations—a result that pushes well beyond the capabilities of existing methods.

PDF Details DOI

YNIMG Journal 2026 Journal Article

Detecting early brain susceptibility changes before demyelination in cuprizone mouse model using quantitative susceptibility mapping (QSM)

Xinyue Han
Jie Chen
Zhuoheng Liu
Juan Liu
Mingquan Lin
Nian Wang

Multiple sclerosis (MS) is a neurological disease that affects the central nervous system through demyelination and inflammation. Animal model, including the cuprizone (CPZ) model, provides a robust platform for studying demyelination and remyelination in MS. While conventional MRI techniques are sensitive to myelin changes, quantitative susceptibility mapping (QSM) offers additional advantages by capturing both myelin- and iron-related pathology. In this study, we performed longitudinal whole-brain multimodal magnetic resonance imaging (MRI), including T2-weighted imaging, magnetization transfer imaging, and QSM, in CPZ-treated mice across multiple stages covering pre-demyelination, acute demyelination, chronic demyelination, and remyelination. Regional analyses focused on the corpus callosum (CC) and anterior commissure (AC), complemented by histological validation. All three MRI modalities detected demyelination, characterized by increased T2 signal, decreased magnetization transfer ratio (MTR), and increased susceptibility, with partial recovery during remyelination. QSM demonstrated unique sensitivity by identifying susceptibility decreases at week 2, before apparent demyelination, corresponding to early oligodendrocyte dysfunction. Regional heterogeneity was observed, with the CC showing rapid alterations during acute demyelination and the AC exhibiting steadier changes across acute and chronic phases. These results establish QSM as a sensitive imaging biomarker capable of detecting early MS pathology and tracking dynamic changes in oligodendrocytes. By complementing conventional MRI techniques, QSM enhances the characterization of white matter injury in the CPZ model and holds translational potential for monitoring disease progression and therapeutic response in MS.

Details DOI

AAAI Conference 2026 Conference Paper

Domain-Aware Multi-View Contrastive Representation Learning for Protein Subcellular Localization Prediction

Qiang Zhang
Feng Yang
Weihong Huang
Jing Feng
Juan Liu

Protein subcellular localization prediction is essential for understanding protein function and cellular organization. However, existing methods exhibit two major limitations: (1) they overlook the critical role of evolutionarily conserved protein domains, which are fundamental functional and structural units that significantly influence functions and subcellular localization, and (2) they rarely learn residue order and backbone coordinates simultaneously, neglecting the complementary information inherent in multi-modal representations. In this paper, we propose a novel Domain-Aware Multi-View Contrastive Representation Learning for Protein Subcellular Localization prediction, named DMVCL. Firstly, it devises domain-sequence/structure attention modules, which identify functionally significant regions in protein structures/sequences that critically determine subcellular localization. Secondly, it introduces a multi-view contrastive learning framework that unites inter-view and intra-view objectives. Inter-view contrastive learning aligns protein sequences with their corresponding structures by maximizing mutual information, thereby capturing the consistency of protein residue order and backbone coordinates. Intra-view contrastive learning enhances the representation discriminability of each modality by explicitly separating proteins with no common location and attracting those with any shared localization. Extensive experiments demonstrate that DMVCL significantly outperforms existing baselines. Ablation studies and visualizations further highlight the contributions of domain-sequence/structure attention and multi-view contrastive learning in achieving superior predictive performance.

PDF Details DOI

AAAI Conference 2026 Conference Paper

Dynamic Geometric Equivariant Network for Full-Atom Antibody Design

Weihong Huang
Feng Yang
Qiang Zhang
Juan Liu

Antibody design is critically important in biomedical and therapeutic contexts but remains extremely challenging due to the complexity of antibody sequence–structure relationships and stringent antigen specificity requirements. Traditional computational approaches rely on multi-stage pipelines and often overlook full-atom details (e.g., side-chain conformations) as well as fine-grained geometric features, resulting in limited effectiveness. To overcome these limitations, we propose Dynamic Geometric Equivariant Network (DGENet), an end-to-end full-atom antibody design model that integrates a geometric-kinematic equivariant dynamic optimization module (GK-EDO) with an full-atom E(3)-equivariant message-passing architecture. This framework enables iterative optimization of antibody structures under explicit geometric and kinematic constraints, generating complete antibody structures (including backbone and side chains) and simultaneously jointly optimizing the sequences and 3D structures of the complementarity-determining regions (CDRs). DGENet also introduces a novel virtual anchor docking mechanism that employs an adaptive PNet-Kabsch module to explicitly guide antibody–antigen binding and achieve precise bound conformations. Evaluations on multiple benchmark datasets demonstrate that DGENet exhibits outstanding performance in antibody structure and sequence generation as well as in designing high-affinity antibodies, underscoring its reliability as an advanced antibody design model.

PDF Details DOI

JBHI Journal 2025 Journal Article

Enhancing Recognition of Stereotyped Movements in ASD Children Through Action Pattern Mining and Multi-Channel Fusion

Baiqiao Zhang
Yanran Yuan
Wei Qin
Xiangxian Li
Weiying Liu
Wenxin Yao
Yulong Bian
Juan Liu

Stereotyped movements play a crucial role in diagnosing Autism Spectrum Disorder (ASD). However, recognizing them poses challenges, due to limited data availability and the movements' specificity and varying duration. To support in-depth analysis of ASD children's movements, we constructed the ACSA653 dataset, comprising 653 videos across six classes of stereotyped movements. This dataset surpasses existing ones in both scale and category. To improve the recognition of stereotyped movements, we propose APMFNet, a model that integrates three modules: Visual Motion Learning (VML), Skeleton Relation Mining (SRM), and Multi-channel Fusion (MF). The VML module focuses on extracting spatial and motion information from RGB and optical-flow sequences. The SRM module effectively mines essential motion patterns associated with stereotyped movements through cross-modal graph. The MF module fuses multi-modal information through cross-modality attention to facilitate decision-making. Tested on ACSA653, APMFNet outperforms current state-of-the-art methods, suggesting its potential to identify stable patterns of stereotyped movements in children with ASD.

Details DOI

IROS Conference 2025 Conference Paper

LGNav: Zero-Shot Object Navigation Driven by Language and Pointing Gesture Using Large Vision-Language Models

Weiyi Zhu
Juan Liu
Xinde Li
Zhiwei Lv
Zhehan Yang

In human communication, referring to a specific object within an environment often involves the combination of a pointing gesture to indicate the object’s direction and linguistic descriptions specifying its name and attributes, thereby enabling precise object identification. Inspired by this natural multimodal interaction, we formalize the zero-shot object navigation driven by language and pointing gesture (LGZSON) task, which aims to more closely approximate real-world human-agent communication scenarios. To address this task, we propose LGNav, an open-set, training-free navigation framework. LGNav estimates the pointing gesture direction by extracting human body landmarks and integrates this directional information with depth images to initialize a versatile candidate position map (VCPM). The framework further employs open-vocabulary object detection to identify all potential candidate objects in the environment, projecting them onto the VCPM. Guided by a motion policy derived from the VCPM, LGNav continuously explores the unknown environment, sequentially visits candidate objects, and utilizes a large vision-language model (LVLM) to verify whether each candidate object satisfies the given navigation instruction. Extensive experimental results validate the effectiveness of LGNav, demonstrating its strong performance in the LG-ZSON task. Furthermore, even in the absence of pointing gestures, LGNav achieves competitive results on standard object navigation benchmarks, including the Gibson and HM3D datasets, outperforming a range of strong baseline methods.

Details

AAAI Conference 2025 Conference Paper

LOHA: Direct Graph Spectral Contrastive Learning Between Low-Pass and High-Pass Views

Ziyun Zou
Yinghui Jiang
Lian Shen
Juan Liu
Xiangrong Liu

Spectral Graph Neural Networks effectively handle graphs with different homophily levels, with low-pass filter mining feature smoothness and high-pass filter capturing differences. When these distinct filters could naturally form two opposite views for self-supervised learning, the commonalities between these counterparts for the same node remain unexplored, leading to suboptimal performance. In this paper, a simple yet effective self-supervised contrastive framework, LOHA, is proposed to address this gap. LOHA optimally leverages low-pass and high-pass views by embracing "harmony in diversity". Rather than solely maximizing the difference between these distinct views, which may lead to feature separation, LOHA harmonizes the diversity by treating the propagation of graph signals from both views as a composite feature. Specifically, a novel high-dimensional feature named spectral signal trend is proposed to serve as the basis for the composite feature, which remains relatively unaffected by changing filters and focuses solely on original feature differences. LOHA achieves an average performance improvement of 2.8% over runner-up models on 9 real-world datasets with varying homophily levels. Notably, LOHA even surpasses fully-supervised models on several datasets, which underscores the potential of LOHA in advancing the efficacy of spectral GNNs for diverse graph structures.

PDF Details DOI

IROS Conference 2024 Conference Paper

A Novel Framework for Structure Descriptors-Guided Hand-drawn Floor Plan Reconstruction

Zhentong Zhang
Juan Liu
Xinde Li
Chuanfei Hu
Fir Dunkin
Shaokun Zhang

In the absence of a pre-built indoor map, robot navigation suffers from the limitations of sensors and environments, resulting in decreased efficiency in performing ad-hoc tasks. Given that blueprints are difficult to obtain, an intuitive method is to provide robots with prior knowledge via hand-drawn floor plans. However, due to the inability of robots to directly comprehend hand-drawn styles, the applicability of this method is limited. In this paper, we present a novel framework for hand-drawn floor plan reconstruction that can recognize abstract hand-drawn elements and standardize the reconstruction of hand-drawn floor plans, thereby providing robots with valuable global map information. Specifically, we design a new series of structure descriptors as reconstruction components and employ a deep learning-based model for recognition. Then the standardized results are obtained through the proposed floor plan reconstruction algorithm. To verify the effectiveness of the framework, we conduct experiments on electronic and paper hand-drawn floor plans. Compared with other state-of-the-art methods, our proposed method achieves superior reconstruction results. This work expands the application scenarios for indoor robots, enabling them to quickly comprehend the semantics of complex scenes, thereby enhancing the competitiveness in downstream tasks.

Details

EAAI Journal 2024 Journal Article

Interpretable detector for cervical cytology using self-attention and cell origin group guidance

Peng Jiang
Juan Liu
Jing Feng
Hua Chen
Yuqi Chen
Cheng Li
Baochuan Pang
Dehua Cao

Deep learning has advanced the development of automated cervical cytology, yet limited studies have delved into methods for incorporating medical domain knowledge, and model interpretability has not been thoroughly investigated. To address this issue, this paper proposes a novel, explainable detection method for abnormal cervical cells, called dual-stream self-attention based feature fusion and origin grouping network (DSA-FFOGNet). To encourage the model to focus more on lesion cells and cell nuclei of diagnostic significance, the dual-stream self-attention (DSA) module is introduced to enhance the learning of lesion-specific features. In view of the complex background, cell dense distribution, cell overlap, or clumps existing in the actual cervical cytology images, multi-scale features are extracted and fused by using the path aggregation network (PAN) to enhance the feature representation ability. By integrating biomedical insights regarding cell provenance and formulating an origin grouping loss, DSA-FFOGNet adjusts the penalties for cervical cells originating from different groups, thereby enhancing the optimization of the model training process. To further improve the detection performance, the classification and localization tasks are decoupled via the use of double detection heads. Extensive experiments validate the robustness of the proposed DSA-FFOGNet. The visualization of class activation maps (CAMs) showcases the model’s interpretability. The proposed approach advances the application and development of explainable artificial intelligence (XAI) models in cervical cytology and inspires further research in automated cervical cytology.

Details DOI

IROS Conference 2023 Conference Paper

Energy Constrained Multi-Agent Reinforcement Learning for Coverage Path Planning

Chenyang Zhao 0009
Juan Liu
Suk-Un Yoon
Xinde Li
Heqing Li
Zhentong Zhang

For multi-agent area coverage path planning problem, existing researches regard it as a combination of Traveling Salesman Problem (TSP) and Coverage Path Planning (CPP). However, these approaches have disadvantages of poor observation ability in online phase and high computational cost in offline phase, making it difficult to be applied to energy-constrained Unmanned Aerial Vehicles (UAVs) and adjust strategy dynamically. In this paper, we decompose the task into two sub-problems: multi-agent path planning and sub-region CPP. We model the multi-agent path planning problem as a Collective Markov Decision Process (C-MDP), and design an Energy Constrained Multi-Agent Reinforcement Learning (ECMARL) algorithm based on the centralized training and distributed execution concept. Taking into account energy constraint of UAVs, the UAV propulsion power model is established to measure the energy consumption of UAVs, and load balancing strategy is applied to dynamically allocate target areas for each UAV. If the UAV is under energy-depleted situation, ECMARL can adjust the mission strategy in real time according to environmental information and energy storage conditions of other UAVs. When UAVs reach each sub-region of interest, Back-an-Forth Paths (BFPs) are adopted to solve CPP problem, which can ensure full coverage, optimality and complexity of the sub-problem. Comprehensive theoretical analysis and experiments demonstrate that ECMARL is superior to the traditional offline TSP-CPP strategy in terms of solution quality and computational time, and can effectively deal with the energy-constrained UAVs.

Details

AIIM Journal 2010 Journal Article

Mixture classification model based on clinical markers for breast cancer prognosis

Tao Zeng
Juan Liu

Objective Accurate cancer prognosis prediction is critical to cancer treatment. There have been many prognosis models based on clinical markers, but few of them are satisfied in clinical applications. And with the development of microarray technologies, cancer researchers have discovered many genes as new markers from the gene expression data and have further developed powerful prognosis models based on these so-called genetic biomarkers. However, the application of such biomarkers still suffers from some problems. The first one is there are a great number of genes and a few samples in the gene expression data so that it is difficult to select a unified gene set to establish a stable classifier for prognosis. The second one is that, due to the experimental and technical reasons, there are existing noises and redundancies in gene expression data, which may lead to building a prognosis predictor with poor performance. The last but not the least one is the microarray experiments are so expensive currently that it is hard to obtain abundant samples. Therefore, it is practical to develop prognosis methods mainly based on conventional clinical markers in real cancer treatment applications. This paper aims to establish an accurate classification model for cancer prognosis, in order to make full use of the invaluable information in clinical data, especially which is usually ignored by most of the existing methods when they aim for high prediction accuracies. Methods First, this paper gives the formal description of general classification problem, and presents a novel mixture classification model to make full use of the invaluable information in clinical data, which is similar to the traditional ensemble classification models except for putting strict constraints on the construction of mapping functions to avoid voting process. Then, a two-layer instance of the proposed model, named as MRS (Mixture of Rough set and Support vector machine), is constructed by integrating rough set and support vector machine (SVM) classification methods, in which, the rough set classifier acts as the first layer to identify some singular samples in data, and the SVM classifier acts as the second layer to classify the remaining samples. Finally, MRS is used to make prognosis prediction on two open breast cancer datasets. One dataset, denoted as BRC-1 hereafter, is a high quality, publicly available dataset of 97 breast cancer tumors of node-negative patients. The other, denoted as BRC-2 hereafter, uses baseline human primary breast tumor data from LBL breast cancer cell collection containing 174 samples. Results We have done two experiments on BRC-1 and BRC-2, respectively. In the first experiment, the BRC-1 dataset is divided into train set with 78 patients (34 ones belonging to poor prognosis group and 44 ones belonging to good prognosis group) and test set with 19 patients (12 ones belonging to poor prognosis group and 7 ones belonging to good prognosis). After trained on the train set, the MRS can correctly classify all the 12 patients with poor prognosis, and 6 of 7 patients with good prognosis in the test set. The results are better than previous researches, even better than the 70-gene based biomarkers. And in the second experiment, we construct the classifiers using BRC-2 dataset, and compare MRS with other representative methods in Weka software by 5-fold cross-validation, and comparison results show that MRS has higher prediction accuracy than those methods. Conclusions The proposed mixture classification model can easily integrate methods with different characteristics. It can overcome the shortcomings of traditional voting-based ensemble models and thus can make full use of the information in clinical data. The experimental results illustrate that our implemented MRS classifier can predict the breast cancer prognosis more accurately than previous prognostic methods.

Details DOI

AIIM Journal 2010 Journal Article

Quantitative prediction of MHC-II binding affinity using particle swarm optimization

Wen Zhang
Juan Liu
Yanqing Niu

Objective Helper T-cell epitopes (Th epitopes) are the basic units which activate helper T-cell's immune response, and they are helpful for understanding the immune mechanism and developing vaccines. Peptide and major histocompatibility complex class II (MHC-II) binding is an important prerequisite event for helper T-cell immune response, and the binding peptides are usually recognized as Th epitopes, therefore we can identify Th epitopes by predicting MHC-II binding peptides. Recently, instead of differentiating the peptides as binder or non-binder, researchers are more interested in predicting binding affinities between MHC-II molecules and peptides. Methodology Motivated by the collective search strategy of the particle swarm optimization algorithm (PSO), a method was developed to make the direct prediction of peptide binding affinity. In our paper, PSO was utilized to search for the optimal position-specific scoring matrices (PSSM) from the experimentally derived allele-related peptides, and then the prediction models were constructed based on the matrices. Moreover, we evaluated several factors influencing the binding affinity, including peptide length and flanking residue length, and incorporated them into our models. Results The performance of our models was evaluated on three MHC-II alleles from AntiJen database and 14 MHC-II alleles from IEDB database. When compared to the existing popular quantitative methods such as MHCPred, SVRMHC, ARB and SMM-align, our method can give out better performance in terms of correlation coefficient (r) and area under ROC curve (AUC). In addition, the results demonstrated that the performance of models was further improved by incorporating the global length information, achieving average AUC value of 0. 7534 and average r value of 0. 4707. Conclusions Quantitative prediction of MHC-II binding affinity can be modeled as an optimization problem. Our PSO based method can find the optimal PSSM, which will then be used for identifying the binding cores and scoring the binding affinities of the peptides. The experiment results show that our method is promising for the prediction of MHC-II binding affinity.

Details DOI

ICAPS Conference 2009 Conference Paper

Pervasive Model Adaptation: The Integration of Planning and Information Gathering in Dynamic Production Systems

Juan Liu
Lukas D. Kuhn
Johan de Kleer
Rong Zhou 0001

Model-based planning often presumes a static system model, while in a practice physical system may evolve or drift over time. This paper proposes the idea of pervasive model adaptation in a production system, where the model is dynamically updated using observation of production output. The core idea is the interplay between model adaptation and production planning. We seek plans which simultaneously serve the goals of achieving high productivity for production, and information gathering for model adaptation. We use a modular printing example to illustrate issues such as formulation of the information criterion and search strategy for informative plans. The idea of pervasive adaptation can be further extended to improve long term productivity in production systems.

Details

IROS Conference 2002 Conference Paper

A connectionist model for localization and route learning based on remembrance of perception and action

Juan Liu
Zixing Cai
Xiaobing Zou

This paper proposes a connectionist model to learn a spatial representation of the world based on temporal memory of perceptions and actions of the robot. It is constructed at run-time to merge the experiences and retrieved in later runs to guide the robot to perform the navigation task. A coding strategy is introduced to extract the directional information from the perception sequence, which endows the robot with localization ability. The Temporal Sequence Processing Network (TSPN) transforms routing knowledge learned from robot experiences into temporal characteristics of cell firing and enables the implicit building of a metric map. The navigation system integrating TSPN and a reactive safeguard module performs collision-free navigation, dynamic landmark and heading detection, route learning and path planning in a noisy world, which is tolerant of sensor inaccuracies and unexpected obstacles. The simulation and real world experiments demonstrate the flexibility and robustness of the system.

Details