Jun Hu

AAAI Conference 2026 Conference Paper

DGP: A Dual-Granularity Prompting Framework for Fraud Detection with Graph-Enhanced LLMs

• Yuan Li
• Jun Hu
• Bryan Hooi
• Bingsheng He
• Cheng Chen

Real-world fraud detection applications benefit from graph learning techniques that jointly exploit node features—often rich in textual data—and graph structural information. Recently, Graph-Enhanced LLMs have emerged as a promising graph learning approach that converts graph information into prompts, exploiting LLMs' ability to reason over both textual and structural information. Among them, text-only prompting, which converts graph information into prompts consisting solely of text tokens, offers a solution that relies only on LLM tuning without requiring additional graph-specific encoders. However, text-only prompting struggles on heterogeneous fraud-detection graphs: multi-hop relations expand exponentially with each additional hop, leading to rapidly growing neighborhoods associated with dense textual information. These neighborhoods may overwhelm the model with long, irrelevant content in the prompt and suppress key signals from the target node, thereby degrading performance. To address this challenge, we propose Dual Granularity Prompting (DGP), which mitigates information overload by preserving fine-grained textual details for the target node while summarizing neighbor information into coarse-grained text prompts. DGP introduces tailored summarization strategies for different data modalities—bi-level semantic abstraction for textual fields and statistical aggregation for numerical features—enabling effective compression of verbose neighbor content into concise, informative prompts. Experiments across public and industry datasets demonstrate that DGP operates within a manageable token budget while improving fraud detection performance by up to 6.8% (AUPRC) over state-of-the-art methods, showing the potential of Graph-Enhanced LLMs for fraud detection.

AAAI Conference 2026 Conference Paper

Echoless Label-Based Pre-computation for Memory-Efficient Heterogeneous Graph Learning

• Jun Hu
• Shangheng Chen
• Yufei He
• Yuan Li
• Bryan Hooi
• Bingsheng He

Heterogeneous Graph Neural Networks (HGNNs) are widely used for deep learning on heterogeneous graphs. Typical end-to-end HGNNs require repetitive message passing during training, limiting efficiency for large-scale real-world graphs. Pre-computation-based HGNNs address this by performing message passing only once during preprocessing, collecting neighbor information into regular-shaped tensors, which enables efficient mini-batch training. Label-based pre-computation methods collect neighbors' label information but suffer from training label leakage, where a node's own label information propagates back to itself during multi-hop message passing—the echo effect. Existing mitigation strategies are memory-inefficient on large graphs or suffer from compatibility issues with advanced message passing methods. We propose Echoless Label-based Pre-computation (Echoless-LP), which eliminates training label leakage with Partition-Focused Echoless Propagation (PFEP). PFEP partitions target nodes and performs echoless propagation, where nodes in each partition collect label information only from neighbors in other partitions, avoiding echo while remaining memory-efficient and compatible with any message passing method. We also introduce an Asymmetric Partitioning Scheme (APS) and a PostAdjust mechanism to address information loss from partitioning and distributional shifts across partitions. Experiments on public datasets demonstrate that Echoless-LP achieves superior performance and maintains memory efficiency compared to baselines.

AAAI Conference 2026 Conference Paper

NTSFormer: A Self-Teaching Graph Transformer for Multimodal Isolated Cold-Start Node Classification

• Jun Hu
• Yufei He
• Yuan Li
• Bryan Hooi
• Bingsheng He

Isolated cold-start node classification on multimodal graphs is challenging because such nodes have no edges and often have missing modalities (e.g., absent text or image features). Existing methods address structural isolation by degrading graph learning models to multilayer perceptrons (MLPs) for isolated cold-start inference, using a teacher model (with graph access) to guide the MLP. However, this results in limited model capacity in the student, which is further challenged when modalities are missing. In this paper, we propose Neighbor-to-Self Graph Transformer (NTSFormer), a unified Graph Transformer framework that jointly tackles the isolation and missing-modality issues via a self-teaching paradigm. Specifically, NTSFormer uses a cold-start attention mask to simultaneously make two predictions for each node: a "student" prediction based only on self information (i.e., the node's own features), and a "teacher" prediction incorporating both self and neighbor information. This enables the model to supervise itself without degrading to an MLP, thereby fully leveraging the Transformer's capacity to handle missing modalities. To handle diverse graph information and missing modalities, NTSFormer performs a one-time multimodal graph pre-computation that converts structural and feature data into token sequences, which are then processed by Mixture-of-Experts (MoE) Input Projection and Transformer layers for effective fusion. Experiments on public datasets show that NTSFormer achieves superior performance for multimodal isolated cold-start node classification.

JBHI Journal 2025 Journal Article

Edge-Guided Multi-Scale Frequency Attention Network for Gastrointestinal Cancer Image Segmentation

• Zhiwen Liao
• Qi Wang
• Xinyi Tang
• Han Wang
• Jun Hu
• Pengxiang Su
• Evangelos K. Markakis
• Peng Luo

Image segmentation is a critical technology to improve the accuracy of clinical decisions and treatments in computer-aided diagnostic systems. However, the diverse morphology and fuzzy boundaries of gastrointestinal tumors incur substantial challenges for existing segmentation models, leading to inaccurate feature capture and generating suboptimal results. For solving these problems, we design an edge-guided multi-scale frequency attention network for the gastrointestinal tumor segmentation task, termed EGMFA-Net, which consists of a Kernel Adaptive Enhancement Module (KAEM) and a Frequency-domain Self-attention Module (FDSA). Specifically, KAEM adaptively adjusts the feature extraction kernel based on the morphology of different lesion regions, which enhances the recognition of different morphology regions via a progressive optimization strategy of feature expression. Furthermore, FDSA effectively aggregates multi-scale features in the frequency domain to achieve global receptive fields while preserving more high-frequency details, thereby enhancing adaptability to complex pathological contexts. Extensive experiments on eight medical image benchmark datasets, including SEED, Kvasir, ClinicDB, ColonDB, ETIS, BKAI, CVC-300, and Synapse, show that EGMFA-Net attains state-of-the-art performance over existing methods. Our implementation is available at https://github.com/med-segment/egmfa-net.

ICML Conference 2025 Conference Paper

Effective and Efficient Masked Image Generation Models

• Zebin You
• Jingyang Ou
• Xiaolu Zhang
• Jun Hu
• Jun Zhou 0011
• Chongxuan Li

Although masked image generation models and masked diffusion models are designed with different motivations and objectives, we observe that they can be unified within a single framework. Building upon this insight, we carefully explore the design space of training and sampling, identifying key factors that contribute to both performance and efficiency. Based on the improvements observed during this exploration, we develop our model, referred to as eMIGM. Empirically, eMIGM demonstrates strong performance on ImageNet generation, as measured by Fréchet Inception Distance (FID). In particular, on ImageNet $256\times256$, with similar number of function evaluations (NFEs) and model parameters, eMIGM outperforms the seminal VAR. Moreover, as NFE and model parameters increase, eMIGM achieves performance comparable to the state-of-the-art continuous diffusion model REPA while requiring less than 45% of the NFE. Additionally, on ImageNet $512\times512$, eMIGM outperforms the strong continuous diffusion model EDM2. Code is available at https: //github. com/ML-GSAI/eMIGM.

NeurIPS Conference 2025 Conference Paper

Large Language Diffusion Models

• Shen Nie
• Fengqi Zhu
• Zebin You
• Xiaolu Zhang
• Jingyang Ou
• Jun Hu
• Jun Zhou
• Yankai Lin

The capabilities of large language models (LLMs) are widely regarded as relying on autoregressive models (ARMs). We challenge this notion by introducing LLaDA, a diffusion model trained from scratch under the pre-training and supervised fine-tuning (SFT) paradigm. LLaDA employs a forward data masking process and a reverse generation process, parameterized by a Transformer to predict masked tokens. It provides a principled generative approach for probabilistic inference by optimizing a likelihood lower bound. Across extensive benchmarks on general tasks, math, code, and so on, LLaDA demonstrates strong scalability and performs comparably to our self-constructed ARM baselines. Remarkably, LLaDA 8B is competitive with strong LLMs like LLaMA3 8B in in-context learning and, after SFT, exhibits impressive instruction-following abilities in case studies such as multi-turn dialogue. Moreover, LLaDA addresses the reversal curse, surpassing GPT-4o in a reversal poem completion task. Our findings show the promise of diffusion models for language modeling at scale and challenge the common assumption that core LLM capabilities discussed above inherently depend on ARMs. Project page and codes: \url{https: //ml-gsai. github. io/LLaDA-demo/}.

AAAI Conference 2025 Conference Paper

Modality-Independent Graph Neural Networks with Global Transformers for Multimodal Recommendation

• Jun Hu
• Bryan Hooi
• Bingsheng He
• Yinwei Wei

Multimodal recommendation systems can learn users' preferences from existing user-item interactions as well as the semantics of multimodal data associated with items. Many existing methods model this through a multimodal user-item graph, approaching multimodal recommendation as a graph learning task. Graph Neural Networks (GNNs) have shown promising performance in this domain. Prior research has capitalized on GNNs' capability to capture neighborhood information within certain receptive fields (typically denoted by the number of hops, K) to enrich user and item semantics. We observe that the optimal receptive fields for GNNs can vary across different modalities. In this paper, we propose GNNs with Modality-Independent Receptive Fields, which employ separate GNNs with independent receptive fields for different modalities to enhance performance. Our results indicate that the optimal K for certain modalities on specific datasets can be as low as 1 or 2, which may restrict the GNNs' capacity to capture global information. To address this, we introduce a Sampling-based Global Transformer, which utilizes uniform global sampling to effectively integrate global information for GNNs. We conduct comprehensive experiments that demonstrate the superiority of our approach over existing methods.

ICRA Conference 2025 Conference Paper

Robust 4D Radar-Aided Inertial Navigation for Aerial Vehicles

• Jinwen Zhu
• Jun Hu
• Xudong Zhao
• Xiaoming Lang
• Yinian Mao
• Guoquan Huang 0003

While LiDAR and cameras are becoming ubiquitous for unmanned aerial vehicles (UAVs) but can be ineffective in challenging environments, 4D millimeter-wave (MMW) radars that can provide robust 3D ranging and Doppler velocity measurements are less exploited for aerial navigation. In this paper, we develop an efficient and robust error-state Kalman filter (ESKF)-based radar-inertial navigation for UAVs. The key idea of the proposed approach is the point-to-distribution radar scan matching to provide motion constraints with proper uncertainty qualification, which are used to update the navigation states in a tightly coupled manner, along with the Doppler velocity measurements. Moreover, we propose a robust keyframe-based matching scheme against the prior map (if available) to bound the accumulated navigation errors and thus provide a radar-based global localization solution with high accuracy. Extensive real-world experimental validations have demonstrated that the proposed radar-aided inertial navigation outperforms state-of-the-art methods in both accuracy and robustness.

AAAI Conference 2025 Conference Paper

Understanding Individual Agent Importance in Multi-Agent System via Counterfactual Reasoning

• Jianming Chen
• Yawen Wang
• Junjie Wang
• Xiaofei Xie
• Jun Hu
• Qing Wang
• Fanjiang Xu

Explaining multi-agent systems (MAS) is urgent as these systems become increasingly prevalent in various applications. Previous work has provided explanations for the actions or states of agents, yet falls short in understanding the blackboxed agent’s importance within a MAS and the overall team strategy. To bridge this gap, we propose EMAI, a novel agent-level explanation approach that evaluates the individual agent’s importance. Inspired by counterfactual reasoning, a larger change in reward caused by the randomized action of agent indicates its higher importance. We model it as a MARL problem to capture interactions across agents. Utilizing counterfactual reasoning, EMAI learns the masking agents to identify important agents. Specifically, we define the optimization function to minimize the reward difference before and after action randomization and introduce sparsity constraints to encourage the exploration of more action randomization of agents during training. The experimental results in seven multi-agent tasks demonstrate that EMAI achieves higher fidelity in explanations compared to baselines and provides more effective guidance in practical applications concerning understanding policies, launching attacks, and patching policies.

IJCAI Conference 2024 Conference Paper

Class-Specific Semantic Generation and Reconstruction Learning for Open Set Recognition

• Liu Haoyang
• Yaojin Lin
• Peipei Li
• Jun Hu
• Xuegang Hu

Open set recognition is a crucial research theme for open-environment machine learning. For this problem, a common solution is to learn compact representations of known classes and identify unknown samples by measuring deviations from these known classes. However, the aforementioned methods (1) lack open training consideration, which is detrimental to the fitting of known classes, and (2) recognize unknown classes on an inadequate basis, which limits the accuracy of recognition. In this study, we propose an open reconstruction learning framework that learns a union boundary region of known classes to characterize unknown space. This facilitates the isolation of known space from unknown space to represent known classes compactly and provides a more reliable recognition basis from the perspective of both known and unknown space. Specifically, an adversarial constraint is used to generate class-specific boundary samples. Then, the known classes and approximate unknown space are fitted with manifolds represented by class-specific auto-encoders. Finally, the auto-encoders output the reconstruction error in terms of known and unknown spaces to recognize samples. Extensive experimental results show that the proposed method outperforms existing advanced methods and achieves new stateof-the-art performance. The code is available at https: //github. com/Ashowman98/CSGRL.

ICRA Conference 2024 Conference Paper

Square-Root Inverse Filter-based GNSS-Visual-Inertial Navigation

• Jun Hu
• Xiaoming Lang
• Feng Zhang
• Yinian Mao
• Guoquan Huang 0003

While Global Navigation Satellite System (GNSS) is often used to provide global positioning if available, its intermittency and/or inaccuracy calls for fusion with other sensors. In this paper, we develop a novel GNSS-Visual-Inertial Navigation System (GVINS) that fuses visual, inertial, and raw GNSS measurements within the square-root inverse sliding window filtering (SRI-SWF) framework in a tightly coupled fashion, which thus is termed SRI-GVINS. In particular, for the first time, we deeply fuse the GNSS pseudorange, Doppler shift, single-differenced pseudorange, and double-differenced carrier phase measurements, along with the visual-inertial measurements. Inherited from the SRI-SWF, the proposed SRI-GVINS gains significant numerical stability and computational efficiency over the start-of-the-art methods. Additionally, we propose to use a filter to sequentially initialize the reference frame transformation till converges, rather than collecting measurements for batch optimization. We also perform online calibration of GNSS-IMU extrinsic parameters to mitigate the possible extrinsic parameter degradation. The proposed SRI-GVINS is extensively evaluated on our own collected UAV datasets and the results demonstrate that the proposed method is able to suppress VIO drift in real-time and also show the effectiveness of online GNSS-IMU extrinsic calibration. The experimental validation on the public datasets further reveals that the proposed SRI-GVINS outperforms the state-of-the-art methods in terms of both accuracy and efficiency.

AAAI Conference 2024 System Paper

Tools Identification By On-Board Adaptation of Vision-and-Language Models

• Jun Hu
• Phil Miller
• Michael Lomnitz
• Saurabh Farkya
• Emre Yilmaz
• Aswin Raghavan
• David Zhang
• Michael Piacentino

A robotic workshop assistant has been a long-standing grand challenge for robotics, speech, computer vision, and artificial intelligence (AI) research. We revisit the goal of visual identification of tools from human queries in the current era of Large Vision-and-Language models (like GPT-4). We find that current off-the-shelf models (that are trained on internet images) are unable to overcome the domain shift and unable to identify small, obscure tools in cluttered environments. Furthermore, these models are unable to match tools to their intended purpose or affordances. We present a novel system for online domain adaptation that can be run directly on a small on-board processor. The system uses Hyperdimensional Computing (HD), a fast and efficient neuromorphic method. We adapted CLIP to work with explicit ("I need the hammer") and implicit purpose-driven queries ("Drive these nails"), and even with depth images as input. This demo allows the user to try out various real tools and interact via free-form audio.

TIST Journal 2023 Journal Article

Second-order Confidence Network for Early Classification of Time Series

• Junwei Lv
• Yuqi Chu
• Jun Hu
• Peipei Li
• Xuegang Hu

Time series data are ubiquitous in a variety of disciplines. Early classification of time series, which aims to predict the class label of a time series as early and accurately as possible, is a significant but challenging task in many time-sensitive applications. Existing approaches mainly utilize heuristic stopping rules to capture stopping signals from the prediction results of time series classifiers. However, heuristic stopping rules can only capture obvious stopping signals, which makes these approaches give either correct but late predictions or early but incorrect predictions. To tackle the problem, we propose a novel second-order confidence network for early classification of time series, which can automatically learn to capture implicit stopping signals in early time series in a unified framework. The proposed model leverages deep neural models to capture temporal patterns and outputs second-order confidence to reflect the implicit stopping signals. Specifically, our model exploits the data not only from a time step but also from the probability sequence to capture stopping signals. By combining stopping signals from the classifier output and the second-order confidence, we design a more robust trigger to decide whether or not to request more observations from future time steps. Experimental results show that our approach can achieve superior results in early classification compared to state-of-the-art approaches.

JBHI Journal 2023 Journal Article

Syn_SegNet: A Joint Deep Neural Network for Ultrahigh-Field 7T MRI Synthesis and Hippocampal Subfield Segmentation in Routine 3T MRI

• Xinwei Li
• Linjin Wang
• Hong Liu
• Baoqiang Ma
• Lei Chu
• Xiaoxi Dong
• Debin Zeng
• Tongtong Che

Precise delineation of hippocampus subfields is crucial for the identification and management of various neurological and psychiatric disorders. However, segmenting these subfields automatically in routine 3T MRI is challenging due to their complex morphology and small size, as well as the limited signal contrast and resolution of the 3T images. This research proposes Syn_SegNet, an end-to-end, multitask joint deep neural network that leverages ultrahigh-field 7T MRI synthesis to improve hippocampal subfield segmentation in 3T MRI. Our approach involves two key components. First, we employ a modified Pix2PixGAN as the synthesis model, incorporating self-attention modules, image and feature matching loss, and ROI loss to generate high-quality 7T-like MRI around the hippocampal region. Second, we utilize a variant of 3D-U-Net with multiscale deep supervision as the segmentation subnetwork, incorporating an anatomic weighted cross-entropy loss that capitalizes on prior anatomical knowledge. We evaluate our method on hippocampal subfield segmentation in paired 3T MRI and 7T MRI with seven different anatomical structures. The experimental findings demonstrate that Syn_SegNet's segmentation performance benefits from integrating synthetic 7T data in an online manner and is superior to competing methods. Furthermore, we assess the generalizability of the proposed approach using a publicly accessible 3T MRI dataset. The developed method would be an efficient tool for segmenting hippocampal subfields in routine clinical 3T MRI.

ICRA Conference 2022 Conference Paper

1D-LRF Aided Visual-Inertial Odometry for High-Altitude MAV Flight

• Jiaxin Hu
• Jun Hu
• Yunjun Shen
• Xiaoming Lang
• Bo Zang
• Guoquan Huang 0003
• Yinian Mao

This paper addresses the problem of visual-inertial odometry (VIO) with a downward facing monocular camera when a micro aerial vehicle (MAV) flying at high altitude (over 100 meters). It is important to note that large scene depth causes visual motion constraints significantly less informative than that in near-sighted scenarios as considered in most existing VIO methods. To cope with this challenge, we develop an efficient MSCKF-based VIO algorithm aided by a single 1D laser range finder (LRF), termed LRF-VIO, which runs in real time on an embedded system. The key idea of the proposed LRF-VIO is to fully exploit the limited metric distance information provided by the 1D LRF to disambiguate the scale during visual feature tracking, thus improving the VIO performance at high altitude. Specifically, during the MSCKF visual measurement update, we deliberately constrain the depth of those SLAM features co-planar with the single LRF measuring point. Additionally, delayed initialization of features utilizes the LRF measurements whenever possible, and online extrinsic calibration between the LRF and monocular camera is performed to further improve estimation accuracy and robustness. The proposed LRF-VIO is extensively validated in both indoor and outdoor real-world experiments, outperforming the state-of-the-art methods.

IJCAI Conference 2021 Conference Paper

Dialogue Disentanglement in Software Engineering: How Far are We?

• Ziyou Jiang
• Lin Shi
• Celia Chen
• Jun Hu
• Qing Wang

Despite the valuable information contained in software chat messages, disentangling them into distinct conversations is an essential prerequisite for any in-depth analyses that utilize this information. To provide a better understanding of the current state-of-the-art, we evaluate five popular dialog disentanglement approaches on software-related chat. We find that existing approaches do not perform well on disentangling software-related dialogs that discuss technical and complex topics. Further investigation on how well the existing disentanglement measures reflect human satisfaction shows that existing measures cannot correctly indicate human satisfaction on disentanglement results. Therefore, in this paper, we introduce and evaluate a novel measure, named DLD. Using results of human satisfaction, we further summarize four most frequently appeared bad disentanglement cases on software-related chat to insight future improvements. These cases include (i) Ignoring Interaction Patterns, (ii) Ignoring Contextual Information, (iii) Mixing up Topics, and (iv) Ignoring User Relationships. We believe that our findings provide valuable insights on the effectiveness of existing dialog disentanglement approaches and these findings would promote a better application of dialog disentanglement in software engineering.

AAMAS Conference 2003 Conference Paper

An Agent-based Architecture for Distributed Interfaces and Timed Media in a Storytelling Application

• Jun Hu
• Loe Feijs

UAI Conference 1997 Conference Paper

Learning Belief Networks in Domains with Recursively Embedded Pseudo Independent Submodels

• Jun Hu
• Yang Xiang