arXiv subjects

Di Wu

Publications and source records attributed to Di Wu.

At least 253 records · Page 14Linked to original sources

arXiv2025

FEAT: A Multi-Agent Forensic AI System with Domain-Adapted Large Language Model for Automated Cause-of-Death Analysis

Forensic cause-of-death determination faces systemic challenges, including workforce shortages and diagnostic variability, particularly in high-volume systems like China's medicolegal infrastructure. We introduce FEAT (ForEnsic AgenT), a multi-agent AI framework that automates and standardizes death investigations through a domain-adapted large language model. FEAT's application-oriented architecture integrates: (i) a central Planner for task decomposition, (ii) specialized Local Solvers for evidence analysis, (iii) a Memory & Reflection module for iterative refinement, and (iv) a Global Solver for conclusion synthesis. The system employs tool-augmented reasoning, hierarchical retrieval-augmented generation, forensic-tuned LLMs, and human-in-the-loop feedback to ensure legal and medical validity. In evaluations across diverse Chinese case cohorts, FEAT outperformed state-of-the-art AI systems in both long-form autopsy analyses and concise cause-of-death conclusions. It demonstrated robust generalization across six geographic regions and achieved high expert concordance in blinded validations. Senior pathologists validated FEAT's outputs as comparable to those of human experts, with improved detection of subtle evidentiary nuances. To our knowledge, FEAT is the first LLM-based AI agent system dedicated to forensic medicine, offering scalable, consistent death certification while maintaining expert-level rigor. By integrating AI efficiency with human oversight, this work could advance equitable access to reliable medicolegal services while addressing critical capacity constraints in forensic systems.

cs.AI↗

arXiv2025

Neuro-MoBRE: Exploring Multi-subject Multi-task Intracranial Decoding via Explicit Heterogeneity Resolving

Neurophysiological decoding, fundamental to advancing brain-computer interface (BCI) technologies, has significantly benefited from recent advances in deep learning. However, existing decoding approaches largely remain constrained to single-task scenarios and individual subjects, limiting their broader applicability and generalizability. Efforts towards creating large-scale neurophysiological foundation models have shown promise, but continue to struggle with significant challenges due to pervasive data heterogeneity across subjects and decoding tasks. Simply increasing model parameters and dataset size without explicitly addressing this heterogeneity fails to replicate the scaling successes seen in natural language processing. Here, we introduce the Neural Mixture of Brain Regional Experts (Neuro-MoBRE), a general-purpose decoding framework explicitly designed to manage the ubiquitous data heterogeneity in neurophysiological modeling. Neuro-MoBRE incorporates a brain-regional-temporal embedding mechanism combined with a mixture-of-experts approach, assigning neural signals from distinct brain regions to specialized regional experts on a unified embedding basis, thus explicitly resolving both structural and functional heterogeneity. Additionally, our region-masked autoencoding pre-training strategy further enhances representational consistency among subjects, complemented by a task-disentangled information aggregation method tailored to effectively handle task-specific neural variations. Evaluations conducted on intracranial recordings from 11 subjects across five diverse tasks, including complex language decoding and epileptic seizure diagnosis, demonstrate that Neuro-MoBRE surpasses prior art and exhibits robust generalization for zero-shot decoding on unseen subjects.

eess.SP↗

arXiv2025

Global Optimality in Multi-Flyby Asteroid Trajectory Optimization: Theory and Application Techniques

Designing optimal trajectories for multi-flyby asteroid missions is scientifically critical but technically challenging due to nonlinear dynamics, intermediate constraints, and numerous local optima. This paper establishes a method that approaches global optimality for multi-flyby trajectory optimization under a given sequence. The original optimal control problem with interior-point equality constraints is transformed into a multi-stage decision formulation. This reformulation enables direct application of dynamic programming in lower dimensions, and follows Bellman's principle of optimality. Moreover, the method provides a quantifiable bound on global optima errors introduced by discretization and approximation assumptions, thus ensuring a measure of confidence in the obtained solution. The method accommodates both impulsive and low-thrust maneuver schemes in rendezvous and flyby scenarios. Several computational techniques are introduced to enhance efficiency, including a specialized solution for bi-impulse cases and an adaptive step refinement strategy. The proposed method is validated through three problems: 1) an impulsive variant of the fourth Global Trajectory Optimization competition problem (GTOC4), 2) the GTOC11 problem, and 3) the original low-thrust GTOC4 problem. Each case demonstrates improvements in fuel consumption over the best-known trajectories. These results give evidence of the generality and effectiveness of the proposed method in global trajectory optimization.

math.OC↗

arXiv2025

Families of Transfers from circular low Earth orbit to Distant Prograde Orbit around the Moon

Distant prograde orbits around the Moon exhibit remarkable potential for practical applications such as cislunar surveillance activities and low-energy transfers due to their instability. Previous works on transfers from circular low Earth orbit to distant prograde orbits mainly focused on construction methods based on dynamical structures, lacking a comprehensive analysis of the solution space of this transfer scenario. This paper investigates the solution space and identifies families of transfers from a 167 km circular low Earth orbit to a 1:1 distant prograde orbit. In particular, grid search and trajectory continuation are performed to construct these transfer trajectories. Initial guesses of the transfers are selected in the 1:1 distant prograde orbit through a backward propagation strategy and are then corrected to satisfy specified constraints. Based on the obtained solutions, a linear predictor is derived to predict more feasible solutions and a predictor-corrector continuation method is used to extend the solution space. Twelve transfer families are identified, most of which are new or previously underexplored. The distributions of construction parameters and transfer characteristics of these twelve families are analyzed and discussed, showing which families are applicable to which types of specific practical missions. Comparison between the obtained solution and solution developed by previous works is further performed to imply the effects of the selection of dynamical model on transfer construction.

astro-ph.EP↗

arXiv2025

FedMLAC: Mutual Learning Driven Heterogeneous Federated Audio Classification

Federated Learning (FL) offers a privacy-preserving framework for training audio classification (AC) models across decentralized clients without sharing raw data. However, Federated Audio Classification (FedAC) faces three major challenges: data heterogeneity, model heterogeneity, and data poisoning, which degrade performance in real-world settings. While existing methods often address these issues separately, a unified and robust solution remains underexplored. We propose FedMLAC, a mutual learning-based FL framework that tackles all three challenges simultaneously. Each client maintains a personalized local AC model and a lightweight, globally shared Plug-in model. These models interact via bidirectional knowledge distillation, enabling global knowledge sharing while adapting to local data distributions, thus addressing both data and model heterogeneity. To counter data poisoning, we introduce a Layer-wise Pruning Aggregation (LPA) strategy that filters anomalous Plug-in updates based on parameter deviations during aggregation. Extensive experiments on four diverse audio classification benchmarks, including both speech and non-speech tasks, show that FedMLAC consistently outperforms state-of-the-art baselines in classification accuracy and robustness to noisy data.

cs.SD↗

arXiv2025

Universal thermodynamic topological classes of static black holes in Conformal Killing Gravity

In this study, we develop universal thermodynamic topological classes for the static black holes in the context of the Conformal Killing Gravity. Our findings indicate that the Conformal Killing Gravity significantly reconstructs the thermodynamic properties of both the smallest inner and the largest outer black hole states. Additionally, it considerably alters the thermodynamic stability of black holes across both high-temperature and low-temperature regimes. This analysis shows that different CKG parameter settings will lead to $W^{0+}$ $(λ>0)$ and $W^{1+}$ $(λ<0)$ categories for the charged AdS black hole, the Reissner-Nordstr$\ddot{o}$m black hole in Conformal Killing Gravity is classified into the $W^{0+}$ and $W^{1+}$ categories. Furthermore, we examine the specific scenario where charge is neglected. The study reveals that within the framework of Conformal Killing Gravity, the Schwarzschild black hole similar to the Schwarzschild-AdS black hole, can be classified into the $W^{1-}$ and $W^{0-}$ categories. This work provides key insights into the fundamental nature of quantum gravity theory.

gr-qc↗

arXiv2025

BS-1-to-N: Diffusion-Based Environment-Aware Cross-BS Channel Knowledge Map Generation for Cell-Free Networks

Channel knowledge map (CKM) inference across base stations (BSs) is the key to achieving efficient environmentaware communications. This paper proposes an environmentaware cross-BS CKM inference method called BS-1-to-N based on the generative diffusion model. To this end, we first design the BS location embedding (BSLE) method tailored for cross-BS CKM inference to embed BS location information in the feature vector of CKM. Further, we utilize the cross- and self-attention mechanism for the proposed BS-1-to-N model to respectively learn the relationships between source and target BSs, as well as that among target BSs. Therefore, given the locations of the source and target BSs, together with the source CKMs as control conditions, cross-BS CKM inference can be performed for an arbitrary number of source and target BSs. Specifically, in architectures with massive distributed nodes like cell-free networks, traditional methods of sequentially traversing each BS for CKM construction are prohibitively costly. By contrast, the proposed BS-1-to-N model is able to achieve efficient CKM inference for a target BS at any potential location based on the CKMs of source BSs. This is achieved by exploiting the fact that within a given area, different BSs share the same wireless environment that leads to their respective CKMs. Therefore, similar to multi-view synthesis, CKMs of different BSs are representations of the same wireless environment from different BS locations. By mining the implicit correlation between CKM and BS location based on the wireless environment, the proposed BS-1-to-N method achieves efficient CKM inference across BSs. We provide extensive comparisons of CKM inference between the proposed BS-1-to-N generative model versus benchmarking schemes, and provide one use case study to demonstrate its practical application for the optimization of BS deployment.

eess.SP↗

arXiv2025

FastTrackTr:Towards Fast Multi-Object Tracking with Transformers

Transformer-based multi-object tracking (MOT) methods have captured the attention of many researchers in recent years. However, these models often suffer from slow inference speeds due to their structure or other issues. To address this problem, we revisited the Joint Detection and Tracking (JDT) method by looking back at past approaches. By integrating the original JDT approach with some advanced theories, this paper employs an efficient method of information transfer between frames on the DETR, constructing a fast and novel JDT-type MOT framework: FastTrackTr. Thanks to the superiority of this information transfer method, our approach not only reduces the number of queries required during tracking but also avoids the excessive introduction of network structures, ensuring model simplicity. Experimental results indicate that our method has the potential to achieve real-time tracking and exhibits competitive tracking accuracy across multiple datasets.

cs.CV↗

arXiv2025

Tensor induced gravitational waves

Primordial gravitational waves on small scales are not tightly constrained by current cosmological observations, which allows for the possibility of large amplitudes at small scales. We investigate second-order tensor induced gravitational waves (TIGWs) sourced by primordial gravitational waves and present the corresponding corrections to the total energy density spectrum of gravitational wave. We analyze primordial gravitational waves with large amplitudes generated by various models at small scales. Our results indicate that when primordial gravitational waves on small scales sufficiently dominate the current PTA observations, corrections to the total energy density spectrum from second-order TIGWs may become pronounced in certain frequency bands.

astro-ph.CO↗

arXiv2025

Hybrid-Domain Synergistic Transformer for Hyperspectral Image Denoising

Hyperspectral image denoising faces the challenge of multi-dimensional coupling of spatially non-uniform noise and spectral correlation interference. Existing deep learning methods mostly focus on RGB images and struggle to effectively handle the unique spatial-spectral characteristics and complex noise distributions of hyperspectral images (HSI). This paper proposes an HSI denoising framework, Hybrid-Domain Synergistic Transformer Network (HDST), based on frequency domain enhancement and multiscale modeling, achieving three-dimensional collaborative processing of spatial, frequency and channel domains. The method innovatively integrates three key mechanisms: (1) introducing an FFT preprocessing module with multi-band convolution to extract cross-band correlations and decouple spectral noise components; (2) designing a dynamic cross-domain attention module that adaptively fuses spatial domain texture features and frequency domain noise priors through a learnable gating mechanism; (3) building a hierarchical architecture where shallow layers capture global noise statistics using multiscale atrous convolution, and deep layers achieve detail recovery through frequency domain postprocessing. Experiments on both real and synthetic datasets demonstrate that HDST significantly improves denoising performance while maintaining computational efficiency, validating the effectiveness of the proposed method. This research provides new insights and a universal framework for addressing complex noise coupling issues in HSI and other high-dimensional visual data. The code is available at https://github.com/lhy-cn/HDST-HSIDenoise.

cs.CV↗

arXiv2025

NeuroCLIP: A Multimodal Contrastive Learning Method for rTMS-treated Methamphetamine Addiction Analysis

Methamphetamine dependence poses a significant global health challenge, yet its assessment and the evaluation of treatments like repetitive transcranial magnetic stimulation (rTMS) frequently depend on subjective self-reports, which may introduce uncertainties. While objective neuroimaging modalities such as electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) offer alternatives, their individual limitations and the reliance on conventional, often hand-crafted, feature extraction can compromise the reliability of derived biomarkers. To overcome these limitations, we propose NeuroCLIP, a novel deep learning framework integrating simultaneously recorded EEG and fNIRS data through a progressive learning strategy. This approach offers a robust and trustworthy biomarker for methamphetamine addiction. Validation experiments show that NeuroCLIP significantly improves discriminative capabilities among the methamphetamine-dependent individuals and healthy controls compared to models using either EEG or only fNIRS alone. Furthermore, the proposed framework facilitates objective, brain-based evaluation of rTMS treatment efficacy, demonstrating measurable shifts in neural patterns towards healthy control profiles after treatment. Critically, we establish the trustworthiness of the multimodal data-driven biomarker by showing its strong correlation with psychometrically validated craving scores. These findings suggest that biomarker derived from EEG-fNIRS data via NeuroCLIP offers enhanced robustness and reliability over single-modality approaches, providing a valuable tool for addiction neuroscience research and potentially improving clinical assessments.

eess.SP↗

arXiv2025

RankMixer: Scaling Up Ranking Models in Industrial Recommenders

Recent progress on large language models (LLMs) has spurred interest in scaling up recommendation systems, yet two practical obstacles remain. First, training and serving cost on industrial Recommenders must respect strict latency bounds and high QPS demands. Second, most human-designed feature-crossing modules in ranking models were inherited from the CPU era and fail to exploit modern GPUs, resulting in low Model Flops Utilization (MFU) and poor scalability. We introduce RankMixer, a hardware-aware model design tailored towards a unified and scalable feature-interaction architecture. RankMixer retains the transformer's high parallelism while replacing quadratic self-attention with multi-head token mixing module for higher efficiency. Besides, RankMixer maintains both the modeling for distinct feature subspaces and cross-feature-space interactions with Per-token FFNs. We further extend it to one billion parameters with a Sparse-MoE variant for higher ROI. A dynamic routing strategy is adapted to address the inadequacy and imbalance of experts training. Experiments show RankMixer's superior scaling abilities on a trillion-scale production dataset. By replacing previously diverse handcrafted low-MFU modules with RankMixer, we boost the model MFU from 4.5\% to 45\%, and scale our ranking model parameters by 100x while maintaining roughly the same inference latency. We verify RankMixer's universality with online A/B tests across two core application scenarios (Recommendation and Advertisement). Finally, we launch 1B Dense-Parameters RankMixer for full traffic serving without increasing the serving cost, which improves user active days by 0.3\% and total in-app usage duration by 1.08\%.

cs.IR↗

arXiv2025

Who Owns This Sample: Cross-Client Membership Inference Attack in Federated Graph Neural Networks

Graph-structured data is prevalent in many real-world applications, including social networks, financial systems, and molecular biology. Graph Neural Networks (GNNs) have become the de facto standard for learning from such data due to their strong representation capabilities. As GNNs are increasingly deployed in federated learning (FL) settings to preserve data locality and privacy, new privacy threats arise from the interaction between graph structures and decentralized training. In this paper, we present the first systematic study of cross-client membership inference attacks (CC-MIA) against node classification tasks of federated GNNs (FedGNNs), where a malicious client aims to infer which client owns the given data. Unlike prior centralized-focused work that focuses on whether a sample was included in training, our attack targets sample-to-client attribution, a finer-grained privacy risk unique to federated settings. We design a general attack framework that exploits FedGNNs' aggregation behaviors, gradient updates, and embedding proximity to link samples to their source clients across training rounds. We evaluate our attack across multiple graph datasets under realistic FL setups. Results show that our method achieves high performance on both membership inference and ownership identification. Our findings highlight a new privacy threat in federated graph learning-client identity leakage through structural and model-level cues, motivating the need for attribution-robust GNN design.

cs.LG↗

arXiv2025

Geometric-Perturbation-Robust Cut-Cell Scheme for Two-Material Flows: Exact Pressure-Equilibrium Preservation and Rigorous Analysis

Preserving pressure equilibrium across material interfaces is critical for the stability of compressible multi-material flow simulations, yet most interface-fitted sharp-interface schemes are notoriously sensitive to interface geometry: even slight perturbations of the captured (or tracked) interface can trigger large spurious pressure oscillations. We present a cut-cell method that is geometric-perturbation-robust (GPR) for the compressible two-material flows. By construction, the scheme provably preserves exact interfacial pressure equilibrium in the presence of small interface-position errors. The key is a strict consistency between the conserved variables and the geometric moments (i.e., the integrals of monomials) of every cut cell. We formulate auxiliary transport equations, whose discrete solutions furnish evolved geometric moment, these geometric moments remain perfectly synchronized with the conserved variables -- even on a deforming mesh. Surpassing the classical geometric conservation law, our approach keeps all higher-order geometric moments consistent, thereby eliminating accuracy loss due to geometric mismatches. To prevent the reconstruction step from destroying pressure equilibrium, we introduce the notion of equilibrium-compatible (EC) reconstructions. A carefully designed modification equips any conventional weighted essentially non-oscillatory (WENO) reconstruction with the EC property; we detail a third-order EC multi-resolution WENO (EC-MRWENO) variant. The tight coupling of EC-MRWENO with the evolved moments yields the first cut-cell solver that is simultaneously provably GPR and genuinely high-order: it attains second-order accuracy precisely at material interfaces, while preserving third-order accuracy in smooth regions. Extensive two-dimensional tests confirm the framework's robustness, accuracy and stability under geometric perturbations and topology changes.

math.NA↗

arXiv2025

Deep Koopman Learning of Nonlinear Time-Varying Systems

This paper presents a data-driven approach to approximate the dynamics of a nonlinear time-varying system (NTVS) by a linear time-varying system (LTVS), which is resulted from the Koopman operator and deep neural networks. Analysis of the approximation error between states of the NTVS and the resulting LTVS is presented. Simulations on a representative NTVS show that the proposed method achieves small approximation errors, even when the system changes rapidly. Furthermore, simulations in an example of quadcopters demonstrate the computational efficiency of the proposed approach.

eess.SY↗

arXiv2025

LONGER: Scaling Up Long Sequence Modeling in Industrial Recommenders

Modeling ultra-long user behavior sequences is critical for capturing both long- and short-term preferences in industrial recommender systems. Existing solutions typically rely on two-stage retrieval or indirect modeling paradigms, incuring upstream-downstream inconsistency and computational inefficiency. In this paper, we present LONGER, a Long-sequence Optimized traNsformer for GPU-Efficient Recommenders. LONGER incorporates (i) a global token mechanism for stabilizing attention over long contexts, (ii) a token merge module with lightweight InnerTransformers and hybrid attention strategy to reduce quadratic complexity, and (iii) a series of engineering optimizations, including training with mixed-precision and activation recomputation, KV cache serving, and the fully synchronous model training and serving framework for unified GPU-based dense and sparse parameter updates. LONGER consistently outperforms strong baselines in both offline metrics and online A/B testing in both advertising and e-commerce services at ByteDance, validating its consistent effectiveness and industrial-level scaling laws. Currently, LONGER has been fully deployed at more than 10 influential scenarios at ByteDance, serving billion users.

cs.IR↗

arXiv2025

Extension OL-MDISF: Online Learning from Mix-Typed, Drifted, and Incomplete Streaming Features

Online learning, where feature spaces can change over time, offers a flexible learning paradigm that has attracted considerable attention. However, it still faces three significant challenges. First, the heterogeneity of real-world data streams with mixed feature types presents challenges for traditional parametric modeling. Second, data stream distributions can shift over time, causing an abrupt and substantial decline in model performance. Additionally, the time and cost constraints make it infeasible to label every data instance in a supervised setting. To overcome these challenges, we propose a new algorithm Online Learning from Mix-typed, Drifted, and Incomplete Streaming Features (OL-MDISF), which aims to relax restrictions on both feature types, data distribution, and supervision information. Our approach involves utilizing copula models to create a comprehensive latent space, employing an adaptive sliding window for detecting drift points to ensure model stability, and establishing label proximity information based on geometric structural relationships. To demonstrate the model's efficiency and effectiveness, we provide theoretical analysis and comprehensive experimental results. This extension serves as a standalone technical reference to the original OL-MDISF method. It provides (i) a contextual analysis of OL-MDISF within the broader landscape of online learning, covering recent advances in mixed-type feature modeling, concept drift adaptation, and weak supervision, and (ii) a comprehensive set of experiments across 14 real-world datasets under two types of drift scenarios. These include full CER trends, ablation studies, sensitivity analyses, and temporal ensemble dynamics. We hope this document can serve as a reproducible benchmark and technical resource for researchers working on nonstationary, heterogeneous, and weakly supervised data streams.

cs.LG↗

arXiv2025

Reinforced Reasoning for Embodied Planning

Embodied planning requires agents to make coherent multi-step decisions based on dynamic visual observations and natural language goals. While recent vision-language models (VLMs) excel at static perception tasks, they struggle with the temporal reasoning, spatial understanding, and commonsense grounding needed for planning in interactive environments. In this work, we introduce a reinforcement fine-tuning framework that brings R1-style reasoning enhancement into embodied planning. We first distill a high-quality dataset from a powerful closed-source model and perform supervised fine-tuning (SFT) to equip the model with structured decision-making priors. We then design a rule-based reward function tailored to multi-step action quality and optimize the policy via Generalized Reinforced Preference Optimization (GRPO). Our approach is evaluated on Embench, a recent benchmark for interactive embodied tasks, covering both in-domain and out-of-domain scenarios. Experimental results show that our method significantly outperforms models of similar or larger scale, including GPT-4o-mini and 70B+ open-source baselines, and exhibits strong generalization to unseen environments. This work highlights the potential of reinforcement-driven reasoning to advance long-horizon planning in embodied AI.

cs.AI↗