Searcharxiv⌕ Search

arXiv subjects

Yuhao Zhang

Publications and source records attributed to Yuhao Zhang.

At least 55 records · Page 3Linked to original sources
arXiv2026

SPRI: SVD-Partitioned Residual Initialization for Data-Constrained MoE Upcycling

Mixture-of-Experts (MoE) models enable efficient scaling, but training them from scratch remains prohibitively expensive. MoE upcycling mitigates this cost by converting pretrained dense models into sparse MoE models. However, existing upcycling methods typically rely on large-scale continued training and often perform poorly under data-constrained supervised adaptation, due to either homogeneous experts or overly disruptive perturbations to pretrained parameters. In this setting, effective upcycling must leverage pretrained weight structure while introducing sufficient diversity among routed experts. To this end, we propose SVD-Partitioned Residual Initialization (SPRI), which distributes SVD-partitioned residuals derived from pretrained feed-forward network (FFN) weights across routed experts, introducing controlled expert diversity grounded in pretrained spectral structure. We further introduce a two-stage training strategy to improve adaptation stability. We evaluate SPRI on multilingual speech-to-text translation, where limited supervised data challenges MoE upcycling and multiple target languages provide natural routing heterogeneity. On CoVoST2 across 15 En-to-XX directions, SPRI improves average BLEU and COMET over fully fine-tuned dense models by 2.58 and 3.32 points, respectively, and outperforms the prior best MoE upcycling baseline by 3.39 BLEU and 4.34 COMET points.

cs.LG↗
arXiv2026

SHAPE: Coalition-Aware Expert Pruning for Sparse Mixture-of-Experts LLMs

Sparse Mixture-of-Experts (MoE) large language models achieve strong quality with low per-token compute, yet their deployment is often limited by the memory wall: the full expert pool must remain resident to support token-dependent routing. Expert pruning is a direct remedy, but prior criteria typically score experts independently and overlook that MoE inference is inherently \emph{coalitional}, where outputs arise from routed top-$k$ expert combinations. We propose \textbf{SHAPE}, a task-driven pruning framework that explicitly models \emph{intra-layer} expert cooperation. SHAPE formulates routing traces on a small calibration set as an empirical cooperative game and assigns interaction-aware expert values via a Shapley-style attribution over observed top-$k$ coalitions, enabling the identification of experts that are essential for high-utility collaborations rather than merely frequent. To preserve MoE topology under a global pruning budget, SHAPE further introduces a \emph{quality-coverage} selection rule that retains, in each layer, the minimal expert subset covering an $α$ fraction of non-negative Shapley mass, while using bisection to match a target keep rate. Experiments on three modern MoE backbones (Qwen3-30B-A3B, GPT-OSS-20B, and DeepSeek-V2-Lite) across diverse benchmarks show that SHAPE consistently improves robustness over global and layer-wise pruning variants, maintaining competitive accuracy under 20\% and 40\% expert pruning without additional training and delivering clear reductions in peak GPU memory footprint. The open-source code is available at https://github.com/Alizen-1009/Shapley-Moe.

cs.LG↗
arXiv2026

Implicit Drifting Policy: One-Step Action Generation via Conditional Expert Geometry

Generative action policies based on diffusion or flow matching excel in behavior cloning, yet their iterative sampling is prohibitive for high-frequency robot control. While recent one-step formulations alleviate this latency, they inevitably discard the intermediate trajectory evolution that provides crucial action correction. Directly recovering this mechanism by explicitly estimating a training-time drifting field is mathematically ill-posed due to extreme conditional demonstration sparsity. We introduce Implicit Drifting Policy (IDP), a one-step imitation learning framework that brings the training-time correction of Drifting into policy learning without explicit vector field estimation. IDP extracts a conditional expert geometry from the local variation of observation-similar expert actions, and compares it against a global reference geometry to isolate condition-specific constraints. This local geometric structure adaptively weights a scalar potential objective. Combined with an expert-proximal terminal evaluation, IDP directly enforces manifold constraints on the one-step generator during training. Extensive evaluations across 2D, 3D, and real-world manipulation tasks show IDP effectively maintains adherence to valid action manifolds, improving upon explicit drifting methods and achieving competitive performance with strong one-step baselines.

cs.RO↗
arXiv2026

EvA: An Evidence-First Audio Understanding Paradigm for LALMs

Large Audio Language Models (LALMs) still struggle in complex acoustic scenes because they often fail to preserve task-relevant acoustic evidence before reasoning begins. We identify this error pattern as the evidence bottleneck: state-of-the-art systems show larger deficits in acoustic evidence extraction than in downstream reasoning, suggesting that upstream perception is often the limiting factor. To address this problem, we propose EvA (Evidence-First Audio), a dual-path architecture that enhances acoustic evidence preservation through hierarchical aggregation and non-compressive, time-aligned fusion. We also build EvA-Perception, a large-scale training set with about 54K event-ordered captions and 500K evidence-grounded QA pairs. Under a unified zero-shot protocol, EvA achieves the best open-source \emph{Perception} results on MMAU, MMAR, and MMSU, with the largest gains on perception-heavy splits. Human evaluation on open-ended captioning further shows improved fine-grained acoustic coverage and caption quality. These results support the evidence-first hypothesis: stronger audio understanding depends on preserving acoustic evidence before reasoning. Project can be found at https://satsuki2486441738.github.io/EvA/.

cs.SD↗
arXiv2026

Ti-Audio: The First Multi-Dialectal End-to-End Speech LLM for Tibetan

Recent advances in Speech Large Language Models (Speech-LLMs) have made significant progress, greatly enhancing multimodal interaction capabilities.However, their application in low-resource and dialect-diverse environments still faces challenges. The severe scarcity of Tibetan data, coupled with the phonetic differences among its major dialects (Ü-Tsang, Amdo, and Kham), is a prime example of this challenge. This paper proposes Ti-Audio, the first multi-dialectal end-to-end Speech-LLM for Tibetan. To efficiently align speech and text, we introduce a Dynamic Q-Former Adapter that extracts essential acoustic features from variable-length speech, ensuring stable cross-modal alignment even with limited data. At the data level, we leverage mutual assistance among related dialects to alleviate data scarcity and employ a temperature-based sampling strategy to maximize this synergy. Experimental results demonstrate that Ti-Audio achieves state-of-the-art performance on Tibetan benchmarks for automatic speech recognition and speech translation. Our work validates the effectiveness of cross-dialectal cooperation and provides a scalable paradigm for the development of Speech-LLM in low-resource scenarios.

cs.SD↗
arXiv2026

SafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Constrained Learning

Vision-language-action models (VLAs) show potential as generalist robot policies. However, these models pose extreme safety challenges during real-world deployment, including the risk of harm to the environment, the robot itself, and humans. How can safety constraints be explicitly integrated into VLAs? We address this by exploring an integrated safety approach (ISA), systematically modeling safety requirements, then actively eliciting diverse unsafe behaviors, effectively constraining VLA policies via safe reinforcement learning, and rigorously assuring their safety through targeted evaluations. Leveraging the constrained Markov decision process (CMDP) paradigm, ISA optimizes VLAs from a min-max perspective against elicited safety risks. Thus, policies aligned through this comprehensive approach achieve the following key features: (I) effective safety-performance trade-offs, reducing the cumulative cost of safety violations by 83.58% compared to the state-of-the-art method, while also maintaining task success rate (+3.85%). (II) strong safety assurance, with the ability to mitigate long-tail risks and handle extreme failure scenarios. (III) robust generalization of learned safety behaviors to various out-of-distribution perturbations. The effectiveness is evaluated on long-horizon mobile manipulation tasks. Our data, models and newly proposed benchmark environment are available at https://pku-safevla.github.io.

cs.RO↗
arXiv2026

Bayesian Optimization for Mixed-Variable Problems in the Natural Sciences

Optimizing expensive black-box objectives over mixed search spaces is a common challenge across the natural sciences. Bayesian optimization (BO) offers sample-efficient strategies through probabilistic surrogate models and acquisition functions. However, its effectiveness diminishes in mixed or high-cardinality discrete spaces, where gradients are unavailable and optimizing the acquisition function becomes computationally demanding. In this work, we generalize the probabilistic reparameterization (PR) approach of Daulton et al. to handle non-equidistant discrete variables, enabling gradient-based optimization in fully mixed-variable settings with Gaussian process (GP) surrogates. With real-world scientific optimization tasks in mind, we conduct systematic benchmarks on synthetic and experimental objectives to obtain an optimized kernel formulations and demonstrate the robustness of our generalized PR method. We additionally show that, when combined with a modified BO workflow, our approach can efficiently optimize highly discontinuous and discretized objective landscapes. This work establishes a practical BO framework for addressing fully mixed optimization problems in the natural sciences, and is particularly well suited to autonomous laboratory settings where noise, discretization, and limited data are inherent.

cs.LG↗
arXiv2026

Online Flow Time Minimization: Tight Bounds for Non-Preemptive Algorithms

This paper studies the online scheduling problem of minimizing total flow time for $n$ jobs on $m$ identical machines. A classical $Ω(n)$ lower bound shows that no deterministic single-machine algorithm can beat the trivial greedy, even when $n$ is known in advance. However, this barrier is specific to deterministic algorithms on a single machine, leaving open what randomization, multiple machines, or the kill-and-restart capability can achieve. We give a nearly complete answer. For randomized non-preemptive algorithms, we establish a tight $Θ(\sqrt{n/m})$ competitive ratio, which also improves the best offline approximation to $O(\sqrt{n/m})$. For deterministic non-preemptive algorithms on multiple machines, we prove an $O(n/m^2 + \sqrt{n/m}\log m)$ upper bound and an $Ω(n/m^2 + \sqrt{n/m})$ lower bound. In the kill-and-restart model, we reveal a sharp transition for deterministic algorithms: $Ω(n/\log n)$ for $m = 1$ versus $Θ(\sqrt{n/m})$ for $m \ge 2$; the latter matches the optimal randomized ratio, and we further show that randomization provides no additional power in this model. We also investigate the setting where $n$ is unknown. We prove that no randomized non-preemptive algorithm achieves $o(n)$ on one machine or $o(n/m^2 + \sqrt{n/m})$ on $m$ machines. In contrast, our kill-and-restart algorithm achieves $O(n^α/\sqrt{m})$ for $m \ge 2$, where $α= (\sqrt{5}-1)/2$, breaking the trivial bound without knowledge of $n$.

cs.DS↗
arXiv2026

Stiff-FCS: Single-Cell Stiffness Profiling With Integrated Molecular and Functional Analysis

Cell stiffness is a key determinant of how cells deform, migrate, and adapt to mechanically restrictive environments, yet existing single-cell stiffness assays remain difficult to combine with molecular analysis and downstream functional studies. To address these limitations, we introduce a microfluidic platform, stiffness-based ferrohydrodynamic cell sorting (Stiff-FCS), designed for high-throughput quantification of single-cell stiffness, on-chip molecular analysis, and post-assay cell recovery. Stiff-FCS combines ferrofluid-driven actuation with graded confinement channels to control cell movement, induce deformation, and spatially separate cells based on stiffness. An inverse computational model converts cell position and morphology into quantitative Young's modulus values. We demonstrate stiffness profiling of hundreds to thousands of cells per chip within minutes, same-cell fluorescence-based protein analysis, and recovery of stiffness-defined cells for downstream assays. Across diverse human and mouse cell lines, Lamin A/C showed the most consistent association with stiffness, whereas softer cells exhibited greater migratory capacity than stiffer cells. In a series of human head and neck cancer cell models, Stiff-FCS further resolved a stiff, less migratory subpopulation enriched in a higher-molecular-weight Vimentin state, offering a workflow for linking single-cell stiffness to molecular heterogeneity and cell behavior.

q-bio.CB↗
arXiv2026

R3DP: Real-Time 3D-Aware Policy for Embodied Manipulation

Embodied manipulation requires accurate 3D understanding of objects and their spatial relations to plan and execute contact-rich actions. While large-scale 3D vision models provide strong priors, their computational cost incurs prohibitive latency for real-time control. We propose Real-time 3D-aware Policy (R3DP), which integrates powerful 3D priors into manipulation policies without sacrificing real-time performance. A core innovation of R3DP is the asynchronous fast-slow collaboration module, which seamlessly integrates large-scale 3D priors into the policy without compromising real-time performance. The system maintains real-time efficiency by querying the pre-trained slow system (VGGT) only on sparse key frames, while simultaneously employing a lightweight Temporal Feature Prediction Network (TFPNet) to predict features for all intermediate frames. By leveraging historical data to exploit temporal correlations, TFPNet explicitly improves task success rates through consistent feature estimation. Additionally, to enable more effective multi-view fusion, we introduce a Multi-View Feature Fuser (MVFF) that aggregates features across views by explicitly incorporating camera intrinsics and extrinsics. R3DP offers a plug-and-play solution for integrating large models into real-time inference systems. We evaluate R3DP against multiple baselines across different visual configurations. R3DP effectively harnesses large-scale 3D priors to achieve superior results, outperforming single-view and multi-view DP by 32.9% and 51.4% in average success rate, respectively. Furthermore, by decoupling heavy 3D reasoning from policy execution, R3DP achieves a 44.8% reduction in inference time compared to a naive DP+VGGT integration.

cs.RO↗
arXiv2026

Approximate Butterfly Counting in Sublinear Time

Bipartite graphs serve as a natural model for representing relationships between two different types of entities. When analyzing bipartite graphs, butterfly counting is a fundamental research problem that aims to count the number of butterflies (i.e., 2x2 bicliques) in a given bipartite graph. While this problem has been extensively studied in the literature, existing algorithms usually necessitate access to a large portion of the entire graph, presenting challenges in real scenarios where graphs are extremely large and I/O costs are expensive. In this paper, we study the butterfly counting problem under the query model, where the following query operations are permitted: degree query, neighbor query, and vertex-pair query. We propose TLS, a practical two-level sampling algorithm that can estimate the butterfly count accurately while accessing only a limited graph structure, achieving significantly lower query costs under the standard query model. TLS also incorporates several key techniques to control the variance, including "small-degree-first sampling" and "wedge sampling via small subsets". To ensure theoretical guarantees, we further introduce two novel techniques: "heavy-light partition" and "guess-and-prove", integrated into TLS. With these techniques, we prove that the algorithm can achieve a (1+eps) accuracy for any given approximation parameter 0 < eps < 1 on general bipartite graphs with a promised time and query complexity. In particular, the promised time is sublinear when the input graph is dense enough. Extensive experiments on 15 datasets demonstrate that TLS delivers robust estimates with up to three orders of magnitude lower query costs and runtime compared to existing solutions.

cs.DB↗
arXiv2026

PAT: Accelerating LLM Decoding via Prefix-Aware Attention with Resource Efficient Multi-Tile Kernel

LLM serving is increasingly dominated by decode attention, which is a memory-bound operation due to massive KV cache loading from global memory. Meanwhile, real-world workloads exhibit substantial, hierarchical shared prefixes across requests (e.g., system prompts, tools/templates, RAG). Existing attention implementations fail to fully exploit prefix sharing: one-query-per-CTA execution repeatedly loads shared prefix KV cache, while one-size-fits-all tiling leaves on-chip resources idle and exacerbates bubbles for uneven KV lengths. These choices amplify memory bandwidth pressure and stall memory-bound decode attention. This paper introduces PAT, a prefix-aware attention kernel implementation for LLM decoding that organizes execution with a pack-forward-merge paradigm. PAT packs queries by shared prefix to reduce repeated memory accesses, runs a customized multi-tile kernel to achieve high resource efficiency. It further applies practical multi-stream forwarding and KV splitting to reduce resource bubbles. The final merge performs online softmax with negligible overhead. We implement PAT as an off-the-shelf plugin for vLLM. Evaluation on both real-world and synthetic workloads shows that PAT reduces attention latency by 53.5% on average and TPOT by 17.0-93.1% under the same configurations against state-of-the-art attention kernels. PAT's source code is publicly available at https://github.com/flashserve/PAT.

cs.DC↗
arXiv2026

Forward and Backward Reachability Analysis of Closed-loop Recurrent Neural Networks via Hybrid Zonotopes

Recurrent neural networks (RNNs) are widely employed to model complex dynamical systems due to their hidden-state structure, which inherently captures temporal dependencies. This work presents a hybrid zonotope-based approach for computing exact forward and backward reachable sets of closed-loop RNN systems with ReLU activation functions. The method formulates state-pair sets to compute reachable sets as hybrid zonotopes without requiring unrolling. To improve scalability, a tunable relaxation scheme is proposed that ranks unstable ReLU units across all layers using a triangle-area score and selectively applies convex relaxations within a fixed binary limit in the hybrid zonotopes. This scheme enables an explicit tradeoff between computational complexity and approximation accuracy, with exact reachability as a special case. In addition, a sufficient condition is derived to certify the safety of closed-loop RNN systems. Numerical examples demonstrate the effectiveness of the proposed approach.

math.OC↗
arXiv2026

A Retrieval-Assisted Framework for Wireless Localization

Accurate and robust wireless localization is a key enabler for a wide range of mobile computing applications. Fingerprint-based localization using channel state information (CSI) has attracted significant attention due to its high accuracy and compatibility with existing communication infrastructures. However, traditional similarity-based fingerprinting methods suffer from high computational complexity and limited scalability in high-dimensional CSI spaces, while purely learning-based approaches fail to explicitly exploit correlations among reference fingerprints during inference. To address these challenges, this paper proposes a unified retrieval-assisted fingerprinting localization framework that tightly integrates similarity-based and learning-based paradigms. Specifically, channel charting is employed to project high-dimensional CSI into a low-dimensional latent space, enabling efficient and scalable retrieval of locally correlated reference points (RPs). Building upon the retrieved RPs, a graph attention network (GAT) is designed to explicitly model inter-sample correlations between the query CSI and its associated references, allowing adaptive and geometry-aware feature aggregation for accurate position estimation. Extensive experiments conducted on both real-world indoor and ray-tracing simulated outdoor scenarios demonstrate that the proposed method consistently outperforms state-of-the-art similarity-based and learning-based localization approaches.

eess.SP↗
arXiv2026

EchoMind: An Interrelated Multi-level Benchmark for Evaluating Empathetic Speech Language Models

Speech Language Models (SLMs) have made significant progress in spoken language understanding. Yet it remains unclear whether they can fully perceive non lexical vocal cues alongside spoken words, and respond with empathy that aligns with both emotional and contextual factors. Existing benchmarks typically evaluate linguistic, acoustic, reasoning, or dialogue abilities in isolation, overlooking the integration of these skills that is crucial for human-like, emotionally intelligent conversation. We present EchoMind, the first interrelated, multi-level benchmark that simulates the cognitive process of empathetic dialogue through sequential, context-linked tasks: spoken-content understanding, vocal-cue perception, integrated reasoning, and response generation. All tasks share identical and semantically neutral scripts that are free of explicit emotional or contextual cues, and controlled variations in vocal style are used to test the effect of delivery independent of the transcript. EchoMind is grounded in an empathy-oriented framework spanning 3 coarse and 12 fine-grained dimensions, encompassing 39 vocal attributes, and evaluated using both objective and subjective metrics. Testing 12 advanced SLMs reveals that even state-of-the-art models struggle with high-expressive vocal cues, limiting empathetic response quality. Analyses of prompt strength, speech source, and ideal vocal cue recognition reveal persistent weaknesses in instruction-following, resilience to natural speech variability, and effective use of vocal cues for empathy. These results underscore the need for SLMs that integrate linguistic content with diverse vocal cues to achieve truly empathetic conversational ability.

cs.CL↗
arXiv2026

OpaqueToolsBench: Learning Nuances of Tool Behavior Through Interaction

Tool-calling is essential for Large Language Model (LLM) agents to complete real-world tasks. While most existing benchmarks assume simple, perfectly documented tools, real-world tools (e.g., general "search" APIs) are often opaque, lacking clear best practices or failure modes. Can LLM agents improve their performance in environments with opaque tools by interacting and subsequently improving documentation? To study this, we create OpaqueToolsBench, a benchmark consisting of three distinct task-oriented environments: general function calling, interactive chess playing, and long-trajectory agentic search. Each environment provides underspecified tools that models must learn to use effectively to complete the task. Results on OpaqueToolsBench suggest existing methods for automatically documenting tools are expensive and unreliable when tools are opaque. To address this, we propose a simple framework, ToolObserver, that iteratively refines tool documentation by observing execution feedback from tool-calling trajectories. Our approach outperforms existing methods on OpaqueToolsBench across datasets, even in relatively hard settings. Furthermore, for test-time tool exploration settings, our method is also efficient, consuming 3.5-7.5x fewer total tokens than the best baseline.

cs.CL↗
arXiv2026

PARD: Enhancing Goodput for Inference Pipeline via Proactive Request Dropping

Modern deep neural network (DNN) applications integrate multiple DNN models into inference pipelines with stringent latency requirements for customized tasks. To mitigate extensive request timeouts caused by accumulation, systems for inference pipelines commonly drop a subset of requests so the remaining ones can satisfy latency constraints. Since it is commonly believed that request dropping adversely affects goodput, existing systems only drop requests when they have to, which we call reactive dropping. However, this reactive policy can not maintain high goodput, as it neither makes timely dropping decisions nor identifies the proper set of requests to drop, leading to issues of dropping requests too late or dropping the wrong set of requests. We propose that the inference system should proactively drop certain requests in advance to enhance the goodput across the entire workload. To achieve this, we design an inference system PARD. It enhances goodput with timely and precise dropping decisions by integrating a proactive dropping method that decides when to drop requests using runtime information of the inference pipeline, and an adaptive request priority mechanism that selects which specific requests to drop based on remaining latency budgets and workload intensity. Evaluation on a cluster of 64 GPUs over real-world workloads shows that PARD achieves $16\%$-$176\%$ higher goodput than the state of the art while reducing the drop rate and wasted computation resources by $1.6\times$-$17\times$ and $1.5\times$-$62\times$ respectively.

cs.DC↗
arXiv2026

Learning with Adaptive Prototype Manifolds for Out-of-Distribution Detection

Out-of-distribution (OOD) detection is a critical task for the safe deployment of machine learning models in the real world. Existing prototype-based representation learning methods have demonstrated exceptional performance. Specifically, we identify two fundamental flaws that universally constrain these methods: the Static Homogeneity Assumption (fixed representational resources for all classes) and the Learning-Inference Disconnect (discarding rich prototype quality knowledge at inference). These flaws fundamentally limit the model's capacity and performance. To address these issues, we propose APEX (Adaptive Prototype for eXtensive OOD Detection), a novel OOD detection framework designed via a Two-Stage Repair process to optimize the learned feature manifold. APEX introduces two key innovations to address these respective flaws: (1) an Adaptive Prototype Manifold (APM), which leverages the Minimum Description Length (MDL) principle to automatically determine the optimal prototype complexity $K_c^*$ for each class, thereby fundamentally resolving prototype collision; and (2) a Posterior-Aware OOD Scoring (PAOS) mechanism, which quantifies prototype quality (cohesion and separation) to bridge the learning-inference disconnect. Comprehensive experiments on benchmarks such as CIFAR-100 validate the superiority of our method, where APEX achieves new state-of-the-art performance.

cs.CV↗