arXiv subjects

Yu Zhu

Publications and source records attributed to Yu Zhu.

At least 55 records · Page 3Linked to original sources

arXiv2025

PoseCrafter: Extreme Pose Estimation with Hybrid Video Synthesis

Pairwise camera pose estimation from sparsely overlapping image pairs remains a critical and unsolved challenge in 3D vision. Most existing methods struggle with image pairs that have small or no overlap. Recent approaches attempt to address this by synthesizing intermediate frames using video interpolation and selecting key frames via a self-consistency score. However, the generated frames are often blurry due to small overlap inputs, and the selection strategies are slow and not explicitly aligned with pose estimation. To solve these cases, we propose Hybrid Video Generation (HVG) to synthesize clearer intermediate frames by coupling a video interpolation model with a pose-conditioned novel view synthesis model, where we also propose a Feature Matching Selector (FMS) based on feature correspondence to select intermediate frames appropriate for pose estimation from the synthesized results. Extensive experiments on Cambridge Landmarks, ScanNet, DL3DV-10K, and NAVI demonstrate that, compared to existing SOTA methods, PoseCrafter can obviously enhance the pose estimation performances, especially on examples with small or no overlap.

cs.CV↗

arXiv2025

Physiology-informed layered sensing for intelligent human-exoskeleton interaction

Wearable exoskeletons hold transformative promise for restoring mobility across diverse users with muscular weakness or other impairments. However, their translation beyond laboratory environments remains limited by sensing systems that capture movement but not underlying physiology. Here, we present a soft, lightweight smart leg sleeve that achieves anatomically aligned, layered multimodal sensing by integrating textile-based surface electromyography (sEMG) electrodes, ultrasensitive textile strain sensors, and inertial measurement units (IMUs). Each sensing modality targets a distinct physiological layer: IMUs track joint kinematics at the skeletal level, sEMG monitors muscle activation at the muscular level, and strain sensors detect skin deformation at the cutaneous level. Together, these sensors provide real-time perception to support three core objectives: controlling personalized assistance, optimizing user effort, and safeguarding against injury risks. The system is skin-conformal, mechanically compliant, and seamlessly integrated with a custom exoskeleton ($<20$~g total sensor and electronics weight). We demonstrate: (1) accurate ankle joint moment estimation (RMSE = 0.13~Nm/kg), (2) real-time classification of metabolic trends (accuracy = 97.1\%), and (3) injury risk detection within 100~ms (recall = 0.96), all validated on unseen users using a leave-one-subject-out protocol. This work establishes a physiology-aligned sensing architecture that reframes exoskeleton perception from motion tracking to real-time physiological decoding, offering a pathway towards intelligent, adaptive, and personalized wearable robotics.

eess.SY↗

arXiv2025

Adaptive Reinforcement Learning for Dynamic Configuration Allocation in Pre-Production Testing

Ensuring reliability in modern software systems requires rigorous pre-production testing across highly heterogeneous and evolving environments. Because exhaustive evaluation is infeasible, practitioners must decide how to allocate limited testing resources across configurations where failure probabilities may drift over time. Existing combinatorial optimization approaches are static, ad hoc, and poorly suited to such non-stationary settings. We introduce a novel reinforcement learning (RL) framework that recasts configuration allocation as a sequential decision-making problem. Our method is the first to integrate Q-learning with a hybrid reward design that fuses simulated outcomes and real-time feedback, enabling both sample efficiency and robustness. In addition, we develop an adaptive online-offline training scheme that allows the agent to quickly track abrupt probability shifts while maintaining long-run stability. Extensive simulation studies demonstrate that our approach consistently outperforms static and optimization-based baselines, approaching oracle performance. This work establishes RL as a powerful new paradigm for adaptive configuration allocation, advancing beyond traditional methods and offering broad applicability to dynamic testing and resource scheduling domains.

cs.SE↗

arXiv2025

An RDMA-First Object Storage System with SmartNIC Offload

AI training and inference impose sustained, fine-grain I/O that stresses host-mediated, TCP-based storage paths. Motivated by kernel-bypass networking and user-space storage stacks, we revisit POSIX-compatible object storage for GPU-centric pipelines. We present ROS2, an RDMA-first object storage system design that offloads the DAOS client to an NVIDIA BlueField-3 SmartNIC while leaving the DAOS I/O engine unchanged on the storage server. ROS2 separates a lightweight control plane (gRPC for namespace and capability exchange) from a high-throughput data plane (UCX/libfabric over RDMA or TCP) and removes host mediation from the data path. Using FIO/DFS across local and remote configurations, we find that on server-grade CPUs RDMA consistently outperforms TCP for both large sequential and small random I/O. When the RDMA-driven DAOS client is offloaded to BlueField-3, end-to-end performance is comparable to the host, demonstrating that SmartNIC offload preserves RDMA efficiency while enabling DPU-resident features such as multi-tenant isolation and inline services (e.g., encryption/decryption) close to the NIC. In contrast, TCP on the SmartNIC lags host performance, underscoring the importance of RDMA for offloaded deployments. Overall, our results indicate that an RDMA-first, SmartNIC-offloaded object-storage stack is a practical foundation for scaling data delivery in modern LLM training environments; integrating optional GPU-direct placement for LLM tasks is left for future work.

cs.AR↗

arXiv2025

Beyond the Binary: The System of All-round Evaluation of Research and Its Practices in China

The lack of a macro-level, systematic evaluation theory to guide the implementation of evaluation practices has become a key bottleneck in the reform of global research evaluation systems. By reviewing the historical development of research evaluation, this paper highlights the current binary opposition between qualitative and quantitative methods in evaluation practices. This paper introduces the System of All-round Evaluation of Research (SAER), a framework that integrates form, content, and utility evaluations with six key elements. SAER offers a theoretical breakthrough by transcending the binary, providing a comprehensive foundation for global evaluation reforms. The comprehensive system proposes a trinity of three evaluation dimensions, combined with six evaluation elements, which would help academic evaluators and researchers reconcile binary oppositions in evaluation methods. The system highlights the dialectical wisdom and experience embedded in Chinese research evaluation theory, offering valuable insights and references for the reform and advancement of global research evaluation systems.

cs.SE↗

arXiv2025

Token-level Accept or Reject: A Micro Alignment Approach for Large Language Models

With the rapid development of Large Language Models (LLMs), aligning these models with human preferences and values is critical to ensuring ethical and safe applications. However, existing alignment techniques such as RLHF or DPO often require direct fine-tuning on LLMs with billions of parameters, resulting in substantial computational costs and inefficiencies. To address this, we propose Micro token-level Accept-Reject Aligning (MARA) approach designed to operate independently of the language models. MARA simplifies the alignment process by decomposing sentence-level preference learning into token-level binary classification, where a compact three-layer fully-connected network determines whether candidate tokens are "Accepted" or "Rejected" as part of the response. Extensive experiments across seven different LLMs and three open-source datasets show that MARA achieves significant improvements in alignment performance while reducing computational costs. The source code and implementation details are publicly available at https://github.com/IAAR-Shanghai/MARA, and the trained models are released at https://huggingface.co/IAAR-Shanghai/MARA_AGENTS.

cs.CL↗

arXiv2025

RoCE BALBOA: Service-enhanced Data Center RDMA for SmartNICs

Data-intensive applications in data centers, especially machine learning (ML), have made the network a bottleneck, which in turn has motivated the development of more efficient network protocols and infrastructure. For instance, remote direct memory access (RDMA) has become the standard protocol for data transport in the cloud as it minimizes data copies and reduces CPU-utilization via host-bypassing. Similarly, an increasing amount of network functions and infrastructure have moved to accelerators, SmartNICs, and in-network computing to bypass the CPU. In this paper we explore the implementation and deployment of RoCE BALBOA, an open-source, RoCE v2-compatible, scalable up to hundreds of queue-pairs, and 100G-capable RDMA-stack that can be used as the basis for building accelerators and smartNICs. RoCE BALBOA is customizable, opening up a design space and offering a degree of adaptability not available in commercial products. We have deployed BALBOA in a cluster using FPGAs and show that it has latency and performance characteristics comparable to commercial NICs. We demonstrate its potential by exploring two classes of use cases. One involves enhancements to the protocol for infrastructure purposes (encryption, deep packet inspection using ML). The other showcases the ability to perform line-rate compute offloads with deep pipelines by implementing commercial data preprocessing pipelines for recommender systems that process the data as it arrives from the network before transferring it directly to the GPU. These examples demonstrate how BALBOA enables the exploration and development of SmartNICs and accelerators operating on network data streams.

cs.AR↗

arXiv2025

Hierarchical Latent Class Models for Mortality Surveillance Using Partially Verified Verbal Autopsies

Monitoring cause-of-death data is an important part of understanding disease burdens and effects of public health interventions. Verbal autopsy (VA) is a well-established method for gathering information about deaths outside of hospitals by conducting an interview to caregivers of a deceased person. It is usually the only tool for cause-of-death surveillance in low-resource settings. A critical limitation with current practices of VA analysis is that all algorithms require either domain knowledge about symptom-cause relationships or large labeled datasets for model training. Therefore, they cannot be easily adopted during public health emergencies when new diseases emerge with rapidly evolving epidemiological patterns. In this paper, we consider estimating the fraction of deaths due to an emerging disease. We develop a novel Bayesian framework using hierarchical latent class models to account for the informative cause-of-death verification process. Our model flexibly captures the joint distribution of symptoms and how they change over time in different sub-populations. We also propose structured priors to improve the precision of the cause-specific mortality estimates for small sub-populations. Our model is motivated by mortality surveillance of COVID-19 related deaths in low-resource settings. We apply our method to a dataset that includes suspected COVID-19 related deaths in Brazil in 2021.

stat.ME↗

arXiv2025

Heterogeneous-IRS-Assisted MIMO Systems: Channel Estimation and Beamforming

Intelligent reflecting surface (IRS) has gained great attention for its ability to create favorable propagation environments. However, the power consumption of conventional IRSs cannot be ignored due to the large number of reflecting elements and control circuits. To balance performance and power consumption, we previously proposed a heterogeneous-IRS (HE-IRS), a green IRS structure integrating dynamically tunable elements (DTEs) and statically tunable elements (STEs). Compared to conventional IRSs with only DTEs, the unique DTE-STE integrated structure introduces new challenges in both channel estimation and beamforming. In this paper, we investigate the channel estimation and beamforming problems in HE-IRS-assisted multi-user multiple-input multiple-output systems. Unlike the overall cascaded channel estimated in conventional IRSs, we show that the HE-IRS channel to be estimated is decomposed into a DTE-based cascaded channel and an STE-based equivalent channel. Leveraging it along with the inherent sparsity of DTE- and STE-based channels and manifold optimization, we propose an efficient channel estimation scheme. To address the rank mismatch problem in the imperfect channel sparsity information, a robust rank selection rule is developed. For beamforming, we propose an offline algorithm to optimize the STE phase shifts for wide beam coverage, and an online algorithm to optimize the BS precoder and the DTE phase shifts using the estimated HE-IRS channel. Simulation results show that the HE-IRS requires less pilot overhead than conventional IRSs with the same number of elements. With the proposed channel estimation and beamforming schemes, the green HE-IRS achieves competitive sum rate performance with significantly reduced power consumption.

eess.SP↗

arXiv2025

Probing the Limit of Heat Transfer in Inorganic Crystals with Deep Learning

Heat transfer is a fundamental property of matter. Research spanning decades has attempted to discover materials with exceptional thermal conductivity, yet the upper limit remains unknown. Using deep learning accelerated crystal structure prediction and first-principles calculation, we systematically explore the thermal conductivity landscape of inorganic crystals. We brute-force over half a million ordered crystalline structures, encompassing an extensive coverage of local energy minima in binary compounds with up to four atoms per primitive cell. We confirm diamond sets the upper bound of thermal conductivity within our search space, very likely also among all stable crystalline solids at ambient conditions. We also identify over 20 novel crystals surpassing silicon in thermal conductivity, validated by density functional theory. These include a semiconductor TaN with ultrahigh thermal conductivity (~900 $\mathrm{W\cdot m^{-1}\cdot K^{-1}}$), and metallic compounds such as MnV that exhibit high lattice and electronic thermal conductivity simultaneously, a distinctive feature not observed before. These results as well as the deep learning-driven screening method, redefine the landscape of thermal transport and establish a large open-access database for future materials discovery.

cond-mat.mtrl-sci↗

arXiv2025

Argus: Federated Non-convex Bilevel Learning over 6G Space-Air-Ground Integrated Network

The space-air-ground integrated network (SAGIN) has recently emerged as a core element in the 6G networks. However, traditional centralized and synchronous optimization algorithms are unsuitable for SAGIN due to infrastructureless and time-varying environments. This paper aims to develop a novel Asynchronous algorithm a.k.a. Argus for tackling non-convex and non-smooth decentralized federated bilevel learning over SAGIN. The proposed algorithm allows networked agents (e.g. autonomous aerial vehicles) to tackle bilevel learning problems in time-varying networks asynchronously, thereby averting stragglers from impeding the overall training speed. We provide a theoretical analysis of the iteration complexity, communication complexity, and computational complexity of Argus. Its effectiveness is further demonstrated through numerical experiments.

cs.LG↗

arXiv2025

Neural Representational Consistency Emerges from Probabilistic Neural-Behavioral Representation Alignment

Individual brains exhibit striking structural and physiological heterogeneity, yet neural circuits can generate remarkably consistent functional properties across individuals, an apparent paradox in neuroscience. While recent studies have observed preserved neural representations in motor cortex through manual alignment across subjects, the zero-shot validation of such preservation and its generalization to more cortices remain unexplored. Here we present PNBA (Probabilistic Neural-Behavioral Representation Alignment), a new framework that leverages probabilistic modeling to address hierarchical variability across trials, sessions, and subjects, with generative constraints preventing representation degeneration. By establishing reliable cross-modal representational alignment, PNBA reveals robust preserved neural representations in monkey primary motor cortex (M1) and dorsal premotor cortex (PMd) through zero-shot validation. We further establish similar representational preservation in mouse primary visual cortex (V1), reflecting a general neural basis. These findings resolve the paradox of neural heterogeneity by establishing zero-shot preserved neural representations across cortices and species, enriching neural coding insights and enabling zero-shot behavior decoding.

q-bio.NC↗

arXiv2025

Sparse2DGS: Geometry-Prioritized Gaussian Splatting for Surface Reconstruction from Sparse Views

We present a Gaussian Splatting method for surface reconstruction using sparse input views. Previous methods relying on dense views struggle with extremely sparse Structure-from-Motion points for initialization. While learning-based Multi-view Stereo (MVS) provides dense 3D points, directly combining it with Gaussian Splatting leads to suboptimal results due to the ill-posed nature of sparse-view geometric optimization. We propose Sparse2DGS, an MVS-initialized Gaussian Splatting pipeline for complete and accurate reconstruction. Our key insight is to incorporate the geometric-prioritized enhancement schemes, allowing for direct and robust geometric learning under ill-posed conditions. Sparse2DGS outperforms existing methods by notable margins while being ${2}\times$ faster than the NeRF-based fine-tuning approach.

cs.CV↗

arXiv2025

Multi-Granularity Language-Guided Training for Multi-Object Tracking

Most existing multi-object tracking methods typically learn visual tracking features via maximizing dis-similarities of different instances and minimizing similarities of the same instance. While such a feature learning scheme achieves promising performance, learning discriminative features solely based on visual information is challenging especially in case of environmental interference such as occlusion, blur and domain variance. In this work, we argue that multi-modal language-driven features provide complementary information to classical visual features, thereby aiding in improving the robustness to such environmental interference. To this end, we propose a new multi-object tracking framework, named LG-MOT, that explicitly leverages language information at different levels of granularity (scene-and instance-level) and combines it with standard visual features to obtain discriminative representations. To develop LG-MOT, we annotate existing MOT datasets with scene-and instance-level language descriptions. We then encode both instance-and scene-level language information into high-dimensional embeddings, which are utilized to guide the visual features during training. At inference, our LG-MOT uses the standard visual features without relying on annotated language descriptions. Extensive experiments on three benchmarks, MOT17, DanceTrack and SportsMOT, reveal the merits of the proposed contributions leading to state-of-the-art performance. On the DanceTrack test set, our LG-MOT achieves an absolute gain of 2.2\% in terms of target object association (IDF1 score), compared to the baseline using only visual features. Further, our LG-MOT exhibits strong cross-domain generalizability. The dataset and code will be available at https://github.com/WesLee88524/LG-MOT.

cs.CV↗

arXiv2025

A spatial hypergraph model to smoothly interpolate between pairwise graphs and hypergraphs to study higher-order structures

We introduce a spatial graph and hypergraph model that smoothly interpolates between a graph with purely pairwise edges and a graph where all connections are in large hyperedges. The key component is a spatial clustering resolution parameter that varies between assigning all the vertices in a spatial region to individual clusters, resulting in the pairwise case, to assigning all the vertices in a spatial region to a single cluster, which results in the large hyperedge case. An important outcome of this model is that the spatial structure is invariant to the choice of hyperedges. Consequently, this model enables us to study clustering coefficients, graph diffusion, and epidemic spread and how their behavior changes as a function of the higher-order structure in the network with a fixed spatial substrate. We hope that our model will find future uses to distill or explain other behaviors in higher-order networks.

cs.SI↗

arXiv2025

5f Electron Induced Spin Transport by Sandwich-Type Phthalocyanine

In this study, we employed the non-equilibrium Green's function method combined with density functional theory to investigate the spin transport properties of the actinide sandwich phthalocyanine molecule U(Pc)2.This study aims to provide beneficial assistance for the development of actinide phthalocyanine molecular spintronic devices.

physics.comp-ph↗

arXiv2025

C-Drag: Chain-of-Thought Driven Motion Controller for Video Generation

Trajectory-based motion control has emerged as an intuitive and efficient approach for controllable video generation. However, the existing trajectory-based approaches are usually limited to only generating the motion trajectory of the controlled object and ignoring the dynamic interactions between the controlled object and its surroundings. To address this limitation, we propose a Chain-of-Thought-based motion controller for controllable video generation, named C-Drag. Instead of directly generating the motion of some objects, our C-Drag first performs object perception and then reasons the dynamic interactions between different objects according to the given motion control of the objects. Specifically, our method includes an object perception module and a Chain-of-Thought-based motion reasoning module. The object perception module employs visual language models to capture the position and category information of various objects within the image. The Chain-of-Thought-based motion reasoning module takes this information as input and conducts a stage-wise reasoning process to generate motion trajectories for each of the affected objects, which are subsequently fed to the diffusion model for video synthesis. Furthermore, we introduce a new video object interaction (VOI) dataset to evaluate the generation quality of motion controlled video generation methods. Our VOI dataset contains three typical types of interactions and provides the motion trajectories of objects that can be used for accurate performance evaluation. Experimental results show that C-Drag achieves promising performance across multiple metrics, excelling in object motion control. Our benchmark, codes, and models will be available at https://github.com/WesLee88524/C-Drag-Official-Repo.

cs.CV↗

arXiv2025

Flexible Bayesian Tensor Decomposition for Verbal Autopsy Data

Cause-of-death data is fundamental for understanding population health trends and inequalities as well as designing and evaluating public health interventions. A significant proportion of global deaths, particularly in low- and middle-income countries (LMICs), do not have medically certified causes assigned. In such settings, verbal autopsy (VA) is a widely adopted approach to estimate disease burdens by interviewing caregivers of the deceased. Recently, latent class models have been developed to model the joint distribution of symptoms and perform probabilistic cause-of-death assignment. A large number of latent classes are usually needed in order to characterize the complex dependence among symptoms, making the estimated symptom profiles challenging to summarize and interpret. In this paper, we propose a flexible Bayesian tensor decomposition framework that balances the predictive accuracy of the cause-of-death assignment task and the interpretability of the latent structures. The key to our approach is to partition symptoms into groups and model the joint distributions of group-level symptom sub-profiles. The proposed methods achieve better predictive accuracy than existing VA methods and provide a more parsimonious representation of the symptom distributions. We show our methods provide new insights into the clustering patterns of both symptoms and causes using the PHMRC gold-standard VA dataset.

stat.AP↗