arXiv subjects

Yuchen Zhang

Publications and source records attributed to Yuchen Zhang.

At least 109 records · Page 6Linked to original sources

arXiv2025

P1: Mastering Physics Olympiads with Reinforcement Learning

Recent progress in large language models (LLMs) has moved the frontier from puzzle-solving to science-grade reasoning-the kind needed to tackle problems whose answers must stand against nature, not merely fit a rubric. Physics is the sharpest test of this shift, which binds symbols to reality in a fundamental way, serving as the cornerstone of most modern technologies. In this work, we manage to advance physics research by developing large language models with exceptional physics reasoning capabilities, especially excel at solving Olympiad-level physics problems. We introduce P1, a family of open-source physics reasoning models trained entirely through reinforcement learning (RL). Among them, P1-235B-A22B is the first open-source model with Gold-medal performance at the latest International Physics Olympiad (IPhO 2025), and wins 12 gold medals out of 13 international/regional physics competitions in 2024/2025. P1-30B-A3B also surpasses almost all other open-source models on IPhO 2025, getting a silver medal. Further equipped with an agentic framework PhysicsMinions, P1-235B-A22B+PhysicsMinions achieves overall No.1 on IPhO 2025, and obtains the highest average score over the 13 physics competitions. Besides physics, P1 models also present great performance on other reasoning tasks like math and coding, showing the great generalibility of P1 series.

cs.LG↗

arXiv2025

Team Xiaomi EV-AD VLA: Caption-Guided Retrieval System for Cross-Modal Drone Navigation -- Technical Report for IROS 2025 RoboSense Challenge Track 4

Cross-modal drone navigation remains a challenging task in robotics, requiring efficient retrieval of relevant images from large-scale databases based on natural language descriptions. The RoboSense 2025 Track 4 challenge addresses this challenge, focusing on robust, natural language-guided cross-view image retrieval across multiple platforms (drones, satellites, and ground cameras). Current baseline methods, while effective for initial retrieval, often struggle to achieve fine-grained semantic matching between text queries and visual content, especially in complex aerial scenes. To address this challenge, we propose a two-stage retrieval refinement method: Caption-Guided Retrieval System (CGRS) that enhances the baseline coarse ranking through intelligent reranking. Our method first leverages a baseline model to obtain an initial coarse ranking of the top 20 most relevant images for each query. We then use Vision-Language-Model (VLM) to generate detailed captions for these candidate images, capturing rich semantic descriptions of their visual content. These generated captions are then used in a multimodal similarity computation framework to perform fine-grained reranking of the original text query, effectively building a semantic bridge between the visual content and natural language descriptions. Our approach significantly improves upon the baseline, achieving a consistent 5\% improvement across all key metrics (Recall@1, Recall@5, and Recall@10). Our approach win TOP-2 in the challenge, demonstrating the practical value of our semantic refinement strategy in real-world robotic navigation scenarios.

cs.RO↗

arXiv2025

Loquetier: A Virtualized Multi-LoRA Framework for Unified LLM Fine-tuning and Serving

Low-Rank Adaptation (LoRA) has become a widely adopted parameter-efficient fine-tuning (PEFT) technique for adapting large language models (LLMs) to downstream tasks. While prior work has explored strategies for integrating LLM training and serving, there still remains a gap in unifying fine-tuning and inference for LoRA-based models. We present Loquetier, a virtualized multi-LoRA framework that seamlessly integrates LoRA fine-tuning and serving within a single runtime. Loquetier introduces two key components: (1) a Virtualized Module that isolates PEFT-based modifications and supports multiple adapters on a shared base model, and (2) an optimized computation flow with a kernel design that merges fine-tuning and inference paths in forward propagation, enabling efficient batching and minimizing kernel invocation overhead. Extensive experiments across three task settings show that Loquetier consistently outperforms existing baselines in both performance and flexibility, achieving up to $3.0\times$ the throughput of the state-of-the-art co-serving system on inference-only tasks and $46.4\times$ higher SLO attainment than PEFT on unified fine-tuning and inference tasks. The implementation of Loquetier is publicly available at https://github.com/NJUDeepEngine/Loquetier.

cs.LG↗

arXiv2025

Sensing Security in Near-Field ISAC: Exploiting Scatterers for Eavesdropper Deception

In this paper, we explore sensing security in near-field (NF) integrated sensing and communication (ISAC) scenarios by exploiting known scatterers in the sensing scene. We propose a location deception (LD) scheme where scatterers are deliberately illuminated with probing power that is higher than that directed toward targets of interest, with the goal of deceiving potential eavesdroppers (Eves) with sensing capability into misidentifying scatterers as targets. While the known scatterers can be removed at the legitimate sensing receiver, our LD approach causes Eves to misdetect targets. Notably, this deception is achieved without requiring any prior information about the Eves' characteristics or locations. To strike a flexible three-way tradeoff among communication, sensing, and sensing-security performance, the sum rate and power allocated to scatterers are weighted and maximized under a legitimate radar signal-to-interference-plus-noise ratio (SINR) constraint. We employ the fractional programming (FP) framework and semidefinite relaxation (SDR) to solve this problem. To evaluate the security of the proposed LD scheme, the Cramer-Rao Bound (CRB) and mean squared error (MSE) metrics are employed. Additionally, we introduce the Kullback-Leibler Divergence (KLD) gap between targets and scatterers at Eve to quantify the impact of the proposed LD framework on Eve's sensing performance from an information-theoretical perspective. Simulation results demonstrate that the proposed LD scheme can flexibly adjust the beamforming strategy according to performance requirements, thereby achieving the desired three-way tradeoff. In particular, in terms of sensing security, the proposed scheme significantly enhances the clutter signal strength at Eve's side, leading to confusion or even missed detection of the actual target.

eess.SP↗

arXiv2025

Enabling Scalable Distributed Beamforming via Networked LEO Satellites Towards 6G

In this paper, we propose scalable distributed beamforming schemes over low Earth orbit (LEO) satellite networks that rely solely on statistical channel state information for downlink orthogonal frequency division multiplexing systems. We begin by introducing the system model and presenting a pragmatic yet effective analog beamformer and user-scheduling design. We then derive a closed-form lower bound on the ergodic sum rate, based on the hardening bound, for the digital beamformer design. Next, we formulate a per-satellite power-constrained sum-rate maximization problem, whose centralized solution, obtained via the weighted minimum mean squared error (WMMSE) framework, establishes performance limits and motivates decentralized strategies. We subsequently introduce two decentralized optimization schemes, based on approximating the hardening bound and decentralizing the WMMSE framework, for representative inter-satellite link topologies. In the Ring scheme, satellites update beamformers locally and exchange intermediate parameters sequentially. In the Star scheme, edge satellites update beamformers locally and in parallel, achieving consensus on intermediate parameters at a central satellite using a penalty-dual decomposition framework. Extensive simulations demonstrate that our distributed designs achieve near-centralized performance with superior scalability, substantially outperforming simple closed-form beamformers and single-satellite baselines in sum rate. Additionally, the delay-overhead trade-off between the two topologies is revealed.

eess.SP↗

arXiv2025

Deep Learning in Dental Image Analysis: A Systematic Review of Datasets, Methodologies, and Emerging Challenges

Efficient analysis and processing of dental images are crucial for dentists to achieve accurate diagnosis and optimal treatment planning. However, dental imaging inherently poses several challenges, such as low contrast, metallic artifacts, and variations in projection angles. Combined with the subjectivity arising from differences in clinicians' expertise, manual interpretation often proves time-consuming and prone to inconsistency. Artificial intelligence (AI)-based automated dental image analysis (DIA) offers a promising solution to these issues and has become an integral part of computer-aided dental diagnosis and treatment. Among various AI technologies, deep learning (DL) stands out as the most widely applied and influential approach due to its superior feature extraction and representation capabilities. To comprehensively summarize recent progress in this field, we focus on the two fundamental aspects of DL research-datasets and models. In this paper, we systematically review 260 studies on DL applications in DIA, including 49 papers on publicly available dental datasets and 211 papers on DL-based algorithms. We first introduce the basic concepts of dental imaging and summarize the characteristics and acquisition methods of existing datasets. Then, we present the foundational techniques of DL and categorize relevant models and algorithms according to different DIA tasks, analyzing their network architectures, optimization strategies, training methods, and performance. Furthermore, we summarize commonly used training and evaluation metrics in the DIA domain. Finally, we discuss the current challenges of existing research and outline potential future directions. We hope that this work provides a valuable and systematic reference for researchers in this field. All supplementary materials and detailed comparison tables will be made publicly available on GitHub.

cs.CV↗

arXiv2025

Adamas: Hadamard Sparse Attention for Efficient Long-Context Inference

Large language models (LLMs) now support context windows of hundreds of thousands to millions of tokens, enabling applications such as long-document summarization, large-scale code synthesis, multi-document question answering and persistent multi-turn dialogue. However, such extended contexts exacerbate the quadratic cost of self-attention, leading to severe latency in autoregressive decoding. Existing sparse attention methods alleviate these costs but rely on heuristic patterns that struggle to recall critical key-value (KV) pairs for each query, resulting in accuracy degradation. We introduce Adamas, a lightweight yet highly accurate sparse attention mechanism designed for long-context inference. Adamas applies the Hadamard transform, bucketization and 2-bit compression to produce compact representations, and leverages Manhattan-distance estimation for efficient top-k selections. Experiments show that Adamas matches the accuracy of full attention with only a 64-token budget, achieves near-lossless performance at 128, and supports up to 8x higher sparsity than prior state-of-the-art (SOTA) methods while delivering up to 4.4x self-attention and 1.5x end-to-end speedups on 32K-length sequences. Remarkably, Adamas attains comparable or even lower perplexity than full attention, underscoring its effectiveness in maintaining accuracy under aggressive sparsity.

cs.CL↗

arXiv2025

Integrated Sensing and Communication with Tri-Hybrid Beamforming Across Electromagnetically Reconfigurable Antennas

Beamforming with a sufficient number of antennas is one of the most significant technologies for both Multi-user (MU) Multiple-input Multiple-output (MIMO) communication and MIMO radar sensing in Integrated Sensing and Communication (ISAC) systems. However, its performance suffers from limited Degrees of Freedom (DoFs) in conventional hybrid beamforming systems. To overcome this, we propose an Electromagnetically Reconfigurable Antenna (ERA)-aided ISAC system, where transmit ERAs dynamically adjust their radiation patterns to enhance system DoFs and improve overall performance. Specifically, we design a tri-hybrid beamforming optimization framework combining digital, analog, and Electromagnetic (EM) beamforming to jointly maximize communication rate and sensing Signal-to-Clutter-plus-Noise Ratio (SCNR). Furthermore, an integrated Fractional Programming (FP) and Manifold Optimization (MO) approach is developed to transform the problem into tractable subproblems with closed-form updates. Simulation results verify that the proposed ERA-ISAC system achieves almost 10 dB Sensing and Communication (S&C) performance gain compared to its conventional hybrid beamforming counterparts with Omnidirectional Antenna (OA).

eess.SP↗

arXiv2025

A Survey of Reinforcement Learning for Large Reasoning Models

In this paper, we survey recent advances in Reinforcement Learning (RL) for reasoning with Large Language Models (LLMs). RL has achieved remarkable success in advancing the frontier of LLM capabilities, particularly in addressing complex logical tasks such as mathematics and coding. As a result, RL has emerged as a foundational methodology for transforming LLMs into LRMs. With the rapid progress of the field, further scaling of RL for LRMs now faces foundational challenges not only in computational resources but also in algorithm design, training data, and infrastructure. To this end, it is timely to revisit the development of this domain, reassess its trajectory, and explore strategies to enhance the scalability of RL toward Artificial SuperIntelligence (ASI). In particular, we examine research applying RL to LLMs and LRMs for reasoning abilities, especially since the release of DeepSeek-R1, including foundational components, core problems, training resources, and downstream applications, to identify future opportunities and directions for this rapidly evolving area. We hope this review will promote future research on RL for broader reasoning models. Github: https://github.com/TsinghuaC3I/Awesome-RL-for-LRMs

cs.CL↗

arXiv2025

Process Reinforcement through Implicit Rewards

Dense process rewards have proven a more effective alternative to the sparse outcome-level rewards in the inference-time scaling of large language models (LLMs), particularly in tasks requiring complex multi-step reasoning. While dense rewards also offer an appealing choice for the reinforcement learning (RL) of LLMs since their fine-grained rewards have the potential to address some inherent issues of outcome rewards, such as training efficiency and credit assignment, this potential remains largely unrealized. This can be primarily attributed to the challenges of training process reward models (PRMs) online, where collecting high-quality process labels is prohibitively expensive, making them particularly vulnerable to reward hacking. To address these challenges, we propose PRIME (Process Reinforcement through IMplicit rEwards), which enables online PRM updates using only policy rollouts and outcome labels through implict process rewards. PRIME combines well with various advantage functions and forgoes the dedicated reward model training phrase that existing approaches require, substantially reducing the development overhead. We demonstrate PRIME's effectiveness on competitional math and coding. Starting from Qwen2.5-Math-7B-Base, PRIME achieves a 15.1% average improvement across several key reasoning benchmarks over the SFT model. Notably, our resulting model, Eurus-2-7B-PRIME, surpasses Qwen2.5-Math-7B-Instruct on seven reasoning benchmarks with 10% of its training data.

cs.LG↗

arXiv2025

Probability density function for dispersion measure of fast radio burst from extragalactic medium

Fast Radio Bursts (FRBs) have emerged as powerful probes in cosmology. An optimized method was recently proposed to extract the cosmic baryon density from localized FRBs by maximizing the joint likelihood function of the extragalactic dispersion measure ($\mathrm{DM}_{\mathrm{ext}}$). In this paper, we identify a crucial factor that was omitted in the probability density function (PDF) for $\mathrm{DM}_{\mathrm{ext}}$ in that method. Using simulated FRB data, we demonstrate that neglecting this factor leads to a systematic bias in the inferred cosmic baryon density, with deviations exceeding the $1σ$ confidence level. This highlights the necessity of including the missing factor for reliable cosmological applications of FRBs. Furthermore, applying our corrected PDF to a sample of 88 real localized FRBs, we find that the baryon density inferred with the original PDF is inconsistent with the Planck 2018 CMB results, whereas our corrected PDF yields excellent agreement.

astro-ph.CO↗

arXiv2025

HiPhO: How Far Are (M)LLMs from Humans in the Latest High School Physics Olympiad Benchmark?

Recently, the physical capabilities of (M)LLMs have garnered increasing attention. However, existing benchmarks for physics suffer from two major gaps: they neither provide systematic and up-to-date coverage of real-world physics competitions such as physics Olympiads, nor enable direct performance comparison with humans. To bridge these gaps, we present HiPhO, the first benchmark dedicated to high school physics Olympiads with human-aligned evaluation. Specifically, HiPhO highlights three key innovations. (1) Comprehensive Data: It compiles 13 latest Olympiad exams from 2024-2025, spanning both international and regional competitions, and covering mixed modalities that encompass problems spanning text-only to diagram-based. (2) Professional Evaluation: We adopt official marking schemes to perform fine-grained grading at both the answer and step level, fully aligned with human examiners to ensure high-quality and domain-specific evaluation. (3) Comparison with Human Contestants: We assign gold, silver, and bronze medals to models based on official medal thresholds, thereby enabling direct comparison between (M)LLMs and human contestants. Our large-scale evaluation of 30 state-of-the-art (M)LLMs shows that: across 13 exams, open-source MLLMs mostly remain at or below the bronze level; open-source LLMs show promising progress with multiple golds; closed-source reasoning MLLMs can achieve 6 to 12 gold medals; and most models still have a significant gap from full marks. These results highlight the performance gap between open-source models and top students, the strong reasoning abilities of closed-source models, and the remaining room for improvement. HiPhO, a human-aligned Olympiad benchmark for multimodal physical reasoning, is open-source at https://github.com/SciYu/HiPhO with a public leaderboard at https://phyarena.github.io/.

cs.AI↗

arXiv2025

SimpleVLA-RL: Scaling VLA Training via Reinforcement Learning

Vision-Language-Action (VLA) models have recently emerged as a powerful paradigm for robotic manipulation. Despite substantial progress enabled by large-scale pretraining and supervised fine-tuning (SFT), these models face two fundamental challenges: (i) the scarcity and high cost of large-scale human-operated robotic trajectories required for SFT scaling, and (ii) limited generalization to tasks involving distribution shift. Recent breakthroughs in Large Reasoning Models (LRMs) demonstrate that reinforcement learning (RL) can dramatically enhance step-by-step reasoning capabilities, raising a natural question: Can RL similarly improve the long-horizon step-by-step action planning of VLA? In this work, we introduce SimpleVLA-RL, an efficient RL framework tailored for VLA models. Building upon veRL, we introduce VLA-specific trajectory sampling, scalable parallelization, multi-environment rendering, and optimized loss computation. When applied to OpenVLA-OFT, SimpleVLA-RL achieves SoTA performance on LIBERO and even outperforms $π_0$ on RoboTwin 1.0\&2.0 with the exploration-enhancing strategies we introduce. SimpleVLA-RL not only reduces dependence on large-scale data and enables robust generalization, but also remarkably surpasses SFT in real-world tasks. Moreover, we identify a novel phenomenon ``pushcut'' during RL training, wherein the policy discovers previously unseen patterns beyond those seen in the previous training process. Github: https://github.com/PRIME-RL/SimpleVLA-RL

cs.RO↗

arXiv2025

A Deep Learning Framework for Joint Channel Acquisition and Communication Optimization in Movable Antenna Systems

This paper presents an end-to-end deep learning framework in a movable antenna (MA)-enabled multiuser communication system. In contrast to the conventional works assuming perfect channel state information (CSI), we address the practical CSI acquisition issue through the design of pilot signals and quantized CSI feedback, and further incorporate the joint optimization of channel estimation, MA placement, and precoding design. The proposed mechanism enables the system to learn an optimized transmission strategy from imperfect channel data, overcoming the limitations of conventional methods that conduct channel estimation and antenna position optimization separately. To balance the performance and overhead, we further extend the proposed framework to optimize the antenna placement based on the statistical CSI. Simulation results demonstrate that the proposed approach consistently outperforms traditional benchmarks in terms of achievable sum-rate of users, especially under limited feedback and sparse channel environments. Notably, it achieves a performance comparable to the widely-adopted gradient-based methods with perfect CSI, while maintaining significantly lower CSI feedback overhead. These results highlight the effectiveness and adaptability of learning-based MA system design for future wireless systems.

cs.IT↗

arXiv2025

Hybrid Codebook Design for Localization Using Electromagnetically Reconfigurable Fluid Antenna System

Electromagnetically reconfigurable fluid antenna systems (ER-FAS) introduce additional degrees of freedom in the electromagnetic (EM) domain by dynamically steering per-antenna radiation patterns, thereby enhancing power efficiency in wireless links. Unlike prior works on spatially reconfigurable FAS, which adjust element positions, ER-FAS provides direct control over each element's EM characteristics to realize on-demand beam-pattern shaping. While existing studies have exploited ER-FAS to boost spectral efficiency, this paper explores its application for downlink localization. We consider a multiple-input single-output (MISO) system in which a multi-antenna ER-FAS at the base station serves a single-antenna user equipment (UE). We consider two reconfigurability paradigms: (i) a synthesis model where each antenna generates desired beampatterns from a finite set of EM basis functions, and (ii) a finite-state selection model in which each antenna selects a pattern from a predefined set of patterns. For both paradigms, we formulate the joint baseband (BB) and EM precoder design to minimize the UE position error bound. In the synthesis case we derive low-dimensional closed-form expressions for both the BB and EM precoders. For the finite-state model we obtain closed-form BB structures and propose a low-complexity block-coordinate-descent algorithm for EM pattern selection. Analytical bounds and extensive simulations show that the proposed hybrid designs for ER-FAS substantially improve UE positioning accuracy over traditional non-reconfigurable arrays.

eess.SP↗

arXiv2025

Symbolic Equation Modeling of Composite Loads: A Kolmogorov-Arnold Network based Learning Approach

With increasing penetration of distributed energy resources installed behind the meter, there is a growing need for adequate modelling of composite loads to enable accurate power system simulation analysis. Existing measurement based load modeling methods either fit fixed-structure physical models, which limits adaptability to evolving load mixes, or employ flexible machine learning methods which are however black boxes and offer limited interpretability. This paper presents a new learning based load modelling method based on Kolmogorov Arnold Networks towards modelling flexibility and interpretability. By actively learning activation functions on edges, KANs automatically derive free form symbolic equations that capture nonlinear relationships among measured variables without prior assumptions about load structure. Case studies demonstrate that the proposed approach outperforms other methods in both accuracy and generalization ability, while uniquely representing composite loads into transparent, interpretable mathematical equations.

eess.SY↗

arXiv2025

SSRL: Self-Search Reinforcement Learning

We investigate the potential of large language models (LLMs) to serve as efficient simulators for agentic search tasks in reinforcement learning (RL), thereby reducing dependence on costly interactions with external search engines. To this end, we first quantify the intrinsic search capability of LLMs via structured prompting and repeated sampling, which we term Self-Search. Our results reveal that LLMs exhibit strong scaling behavior with respect to the inference budget, achieving high pass@k on question-answering benchmarks, including the challenging BrowseComp task. Building on these observations, we introduce Self-Search RL (SSRL), which enhances LLMs' Self-Search capability through format-based and rule-based rewards. SSRL enables models to iteratively refine their knowledge utilization internally, without requiring access to external tools. Empirical evaluations demonstrate that SSRL-trained policy models provide a cost-effective and stable environment for search-driven RL training, reducing reliance on external search engines and facilitating robust sim-to-real transfer. We draw the following conclusions: 1) LLMs possess world knowledge that can be effectively elicited to achieve high performance; 2) SSRL demonstrates the potential of leveraging internal knowledge to reduce hallucination; 3) SSRL-trained models integrate seamlessly with external search engines without additional effort. Our findings highlight the potential of LLMs to support more scalable RL agent training.

cs.CL↗

arXiv2025

Echo: Decoupling Inference and Training for Large-Scale RL Alignment on Heterogeneous Swarms

Modern RL-based post-training for large language models (LLMs) co-locate trajectory sampling and policy optimisation on the same GPU cluster, forcing the system to switch between inference and training workloads. This serial context switching violates the single-program-multiple-data (SPMD) assumption underlying today's distributed training systems. We present Echo, the RL system that cleanly decouples these two phases across heterogeneous "inference" and "training" swarms while preserving statistical efficiency. Echo introduces two lightweight synchronization protocols: a sequential pull mode that refreshes policy weights according to API call for minimal bias, and an asynchronous push-pull mode that streams version-tagged rollouts through a replay buffer to maximise hardware utilisation. Training four representative RL workloads with Qwen3-4B, Qwen2.5-7B, Qwen3-30B-A3B-Thinking-2507 and Qwen3-32B on a geographically distributed cluster, Echo matches a fully co-located Verl baseline in convergence speed and final reward while off-loading trajectory generation to commodity edge hardware. These promising results demonstrate that large-scale RL for LLMs could achieve datacentre-grade performance using decentralised, heterogeneous resources.

cs.LG↗