arXiv subjects

Yi Zhang

Publications and source records attributed to Yi Zhang.

At least 361 records · Page 20Linked to original sources

arXiv2025

RLAX: Large-Scale, Distributed Reinforcement Learning for Large Language Models on TPUs

Reinforcement learning (RL) has emerged as the de-facto paradigm for improving the reasoning capabilities of large language models (LLMs). We have developed RLAX, a scalable RL framework on TPUs. RLAX employs a parameter-server architecture. A master trainer periodically pushes updated model weights to the parameter server while a fleet of inference workers pull the latest weights and generates new rollouts. We introduce a suite of system techniques to enable scalable and preemptible RL for a diverse set of state-of-art RL algorithms. To accelerate convergence and improve model quality, we have devised new dataset curation and alignment techniques. Large-scale evaluations show that RLAX improves QwQ-32B's pass@8 accuracy by 12.8% in just 12 hours 48 minutes on 1024 v5p TPUs, while remaining robust to preemptions during training.

cs.LG↗

arXiv2025

Intelligent Resilience Testing for Decision-Making Agents with Dual-Mode Surrogate Adaptation

Testing and evaluating decision-making agents remains challenging due to unknown system architectures, limited access to internal states, and the vastness of high-dimensional scenario spaces. Existing testing approaches often rely on surrogate models of decision-making agents to generate large-scale scenario libraries; however, discrepancies between surrogate models and real decision-making agents significantly limit their generalizability and practical applicability. To address this challenge, this paper proposes intelligent resilience testing (IRTest), a unified online adaptive testing framework designed to rapidly adjust to diverse decision-making agents. IRTest initializes with an offline-trained surrogate prediction model and progressively reduces surrogate-to-real gap during testing through two complementary adaptation mechanisms: (i) online neural fine-tuning in data-rich regimes, and (ii) lightweight importance-sampling-based weighting correction in data-limited regimes. A Bayesian optimization strategy, equipped with bias-corrected acquisition functions, guides scenario generation to balance exploration and exploitation in complex testing spaces. Extensive experiments across varying levels of task complexity and system heterogeneity demonstrate that IRTest consistently improves failure-discovery efficiency, testing robustness, and cross-system generalizability. These results highlight the potential of IRTest as a practical solution for scalable, adaptive, and resilient testing of decision-making agents.

eess.SY↗

arXiv2025

Training-Free Dual Hyperbolic Adapters for Better Cross-Modal Reasoning

Recent research in Vision-Language Models (VLMs) has significantly advanced our capabilities in cross-modal reasoning. However, existing methods suffer from performance degradation with domain changes or require substantial computational resources for fine-tuning in new domains. To address this issue, we develop a new adaptation method for large vision-language models, called \textit{Training-free Dual Hyperbolic Adapters} (T-DHA). We characterize the vision-language relationship between semantic concepts, which typically has a hierarchical tree structure, in the hyperbolic space instead of the traditional Euclidean space. Hyperbolic spaces exhibit exponential volume growth with radius, unlike the polynomial growth in Euclidean space. We find that this unique property is particularly effective for embedding hierarchical data structures using the Poincaré ball model, achieving significantly improved representation and discrimination power. Coupled with negative learning, it provides more accurate and robust classifications with fewer feature dimensions. Our extensive experimental results on various datasets demonstrate that the T-DHA method significantly outperforms existing state-of-the-art methods in few-shot image recognition and domain generalization tasks.

cs.CV↗

arXiv2025

Long-wavelength UV-LEDs and charge management in the detection of gravitational waves in space

For the charge management system in gravitational wave detection missions, a continuous discharge strategy is considered by continuously illuminating a test mass (TM) with weak light in such a way to strike a balance between the charging and discharging rates and at the same time avoids the requirement for frequent activation of charge measurements. Built on experiments by one of us based on a simple parallel plate model for inertial sensor, in the present work a more sophisticated inertial sensor model that mimics the surface properties and work function of a cubical TM of an inertial sensor in space (like that of the LISA Pathfinder) is employed to study bipolar charge management system that utilizes UV-LEDs with peak wavelengths of 269 nm, 275 nm, 280 nm, and 295 nm that are longer than the standard 255 nm commonly employed for direct TM illumination. Experimental results indicate that the 275 nm UV-LED achieves optimal performance, maintaining the TM potential closer to zero and at the same time accommodates both rapid discharge and continuous discharge strategies. The present work provides useful input in the future study of system design and optimization for the charge management system.

gr-qc↗

arXiv2025

Generalized Connes-Kreimer Hopf algebras on decorated rooted forests by weighted cocycles

The Connes-Kreimer Hopf algebra of rooted trees is an operated Hopf algebra whose coproduct satisfies the classical Hochschild 1-cocycle condition. In this paper, we extend the setting from rooted trees to the space $H_{\rm RT}(X,Ω)$ of $(X,Ω)$-rooted trees, in which internal vertices are decorated by a set $Ω$ and leafs are decorated by $X \cup Ω$. We introduce a new coalgebra structure on $H_{\rm RT}(X,Ω)$ whose coproduct satisfies a weighted Hochschild 1-cocycle condition involving multiple operators, thereby generalizing the classical condition. A combinatorial interpretation of this coproduct is also provided. We then endow $H_{\rm RT}(X,Ω)$ with a Hopf algebra structure. Finally, we define weighted $Ω$-cocycle Hopf algebras, characterized by a Hochschild 1-cocycle condition with weights, and show that $H_{\rm RT}(X,Ω)$ is the free object in the category of $Ω$-cocycle Hopf algebras.

math.QA↗

arXiv2025

Search for Ultralight Dark Matter with Quantum Magnetometry in the Earth's Cavity

Ultralight dark matter candidates, such as axions and dark photons, are leading dark matter candidates. They may couple feebly to photons, sourcing oscillating electromagnetic signals in the Earth's conducting cavity formed between the ground and the ionosphere, providing detectable magnetic field signatures at wavelengths above the Earth's size. We carry out a project aiming to search for new physics using an unshielded high-sensitivity atomic magnetometer, termed the Geomagnetic Probe for nEw physiCS (GPEX). In this work, we report our first search for axion and dark photon dark matter, conducted in the desert of XiaoDushan in Gansu Province, China. Analysis of the collection of one-hour data shows no robust evidence for axion- or dark photon-induced magnetic signals. Correspondingly, we set the constraints on the axion-photon coupling with $g_{aγγ} < 7\times10^{-10}\, \mathrm{GeV^{-1}}$ and the dark photon kinetic-mixing parameter $ε< 2\times10^{-6}$ in the mass range $3.5 \times 10^{-16}\, \mathrm{eV} \sim 1.8 \times 10^{-14}\, \mathrm{eV}$. Our findings demonstrate the feasibility of using ground-based quantum magnetic sensors for ultralight dark matter searches. Future networks of such detectors operating over extended periods could improve the sensitivity by about three orders of magnitude.

hep-ph↗

arXiv2025

TCC-Bench: Benchmarking the Traditional Chinese Culture Understanding Capabilities of MLLMs

Recent progress in Multimodal Large Language Models (MLLMs) have significantly enhanced the ability of artificial intelligence systems to understand and generate multimodal content. However, these models often exhibit limited effectiveness when applied to non-Western cultural contexts, which raises concerns about their wider applicability. To address this limitation, we propose the Traditional Chinese Culture understanding Benchmark (TCC-Bench), a bilingual (i.e., Chinese and English) Visual Question Answering (VQA) benchmark specifically designed for assessing the understanding of traditional Chinese culture by MLLMs. TCC-Bench comprises culturally rich and visually diverse data, incorporating images from museum artifacts, everyday life scenes, comics, and other culturally significant contexts. We adopt a semi-automated pipeline that utilizes GPT-4o in text-only mode to generate candidate questions, followed by human curation to ensure data quality and avoid potential data leakage. The benchmark also avoids language bias by preventing direct disclosure of cultural concepts within question texts. Experimental evaluations across a wide range of MLLMs demonstrate that current models still face significant challenges when reasoning about culturally grounded visual content. The results highlight the need for further research in developing culturally inclusive and context-aware multimodal systems. The code and data can be found at: https://tcc-bench.github.io/.

cs.MM↗

arXiv2025

Do Large Language Models Think Like the Brain? Sentence-Level Evidences from Layer-Wise Embeddings and fMRI

Understanding whether large language models (LLMs) and the human brain converge on similar computational principles remains a fundamental and important question in cognitive neuroscience and AI. Do the brain-like patterns observed in LLMs emerge simply from scaling, or do they reflect deeper alignment with the architecture of human language processing? This study focuses on the sentence-level neural mechanisms of language models, systematically investigating how layer-wise representations in LLMs align with the dynamic neural responses during human sentence comprehension. By comparing hierarchical embeddings from 14 publicly available LLMs with fMRI data collected from participants, who were exposed to a naturalistic narrative story, we constructed sentence-level neural prediction models to identify the model layers most significantly correlated with brain region activations. Results show that improvements in model performance drive the evolution of representational architectures toward brain-like hierarchies, particularly achieving stronger functional and anatomical correspondence at higher semantic abstraction levels. These findings advance our understanding of the computational parallels between LLMs and the human brain, highlighting the potential of LLMs as models for human language processing.

cs.CL↗

arXiv2025

Approaching the Low Optical Loss Limit of Plasmonics using Potassium Metals

Plasmonics enables the miniaturization of photonic devices beyond the optical diffraction limit,yet its potential is hindered by inherently large ohmic losses. Hence, it is prudent to explore low loss alternatives to the current mainstay of plasmonics-the noble metals. In this work, we demonstrate the potential of potassium as a plasmonic material with intrinsically low losses in the optical region. The ultraflat, high quality potassium film, fabricated via a rapid slipping-assisted oxide-free crystallization process, achieves an experimentally observed optical damping rate of 3.7 meV,with a measured imaginary permittivity of approximately 0.1 across the entire visible to near-infrared range (400-2000 nm). Near field optical spectroscopic measurements further confirmed the reduced losses by revealing deeply subwavelength confinement of optical modes. This result enhances our understanding of the factors governing plasmonic materials and devices and establishes a new platform for exploring extreme light-matter interactions in a variety of plasmonic systems.

physics.optics↗

arXiv2025

Fault-Tolerant MARL for CAVs under Observation Perturbations for Highway On-Ramp Merging

Multi-Agent Reinforcement Learning (MARL) holds significant promise for enabling cooperative driving among Connected and Automated Vehicles (CAVs). However, its practical application is hindered by a critical limitation, i.e., insufficient fault tolerance against observational faults. Such faults, which appear as perturbations in the vehicles' perceived data, can substantially compromise the performance of MARL-based driving systems. Addressing this problem presents two primary challenges. One is to generate adversarial perturbations that effectively stress the policy during training, and the other is to equip vehicles with the capability to mitigate the impact of corrupted observations. To overcome the challenges, we propose a fault-tolerant MARL method for cooperative on-ramp vehicles incorporating two key agents. First, an adversarial fault injection agent is co-trained to generate perturbations that actively challenge and harden the vehicle policies. Second, we design a novel fault-tolerant vehicle agent equipped with a self-diagnosis capability, which leverages the inherent spatio-temporal correlations in vehicle state sequences to detect faults and reconstruct credible observations, thereby shielding the policy from misleading inputs. Experiments in a simulated highway merging scenario demonstrate that our method significantly outperforms baseline MARL approaches, achieving near-fault-free levels of safety and efficiency under various observation fault patterns.

cs.RO↗

arXiv2025

Beyond Introspection: Reinforcing Thinking via Externalist Behavioral Feedback

While inference-time thinking allows Large Language Models (LLMs) to address complex problems, the extended thinking process can be unreliable or inconsistent because of the model's probabilistic nature, especially near its knowledge boundaries. Existing approaches attempt to mitigate this by having the model critique its own reasoning to make corrections. However, such self-critique inherits the same biases of the original output, known as the introspection illusion. Moving beyond such introspection and inspired by core methodologies in ethology, we propose an externalist three-step framework Distillation-Reinforcement-Reasoning (DRR). Rather than relying on a model's introspection, DRR evaluates its observable behaviors to provide corrective feedback. DRR first distills the reasoner's behavioral traces, then trains a lightweight, external Discriminative Model (DM). At inference time, this DM acts as a critic, identifying and rejecting suspicious reasoning steps. This external feedback compels the LLM to discard flawed pathways and explore alternatives, thereby enhancing reasoning quality without altering the base model. Experiments on multiple reasoning benchmarks show that our framework significantly outperforms prominent self-critique methods. Benefiting from a lightweight and annotation-free design, DRR offers a scalable and adaptable solution for improving the reliability of reasoning in a wide range of LLMs.

cs.LG↗

arXiv2025

DreamO: A Unified Framework for Image Customization

Recently, extensive research on image customization (e.g., identity, subject, style, background, etc.) demonstrates strong customization capabilities in large-scale generative models. However, most approaches are designed for specific tasks, restricting their generalizability to combine different types of condition. Developing a unified framework for image customization remains an open challenge. In this paper, we present DreamO, an image customization framework designed to support a wide range of tasks while facilitating seamless integration of multiple conditions. Specifically, DreamO utilizes a diffusion transformer (DiT) framework to uniformly process input of different types. During training, we construct a large-scale training dataset that includes various customization tasks, and we introduce a feature routing constraint to facilitate the precise querying of relevant information from reference images. Additionally, we design a placeholder strategy that associates specific placeholders with conditions at particular positions, enabling control over the placement of conditions in the generated results. Moreover, we employ a progressive training strategy consisting of three stages: an initial stage focused on simple tasks with limited data to establish baseline consistency, a full-scale training stage to comprehensively enhance the customization capabilities, and a final quality alignment stage to correct quality biases introduced by low-quality data. Extensive experiments demonstrate that the proposed DreamO can effectively perform various image customization tasks with high quality and flexibly integrate different types of control conditions.

cs.CV↗

arXiv2025

Automated Composition of Agents: A Knapsack Approach for Agentic Component Selection

Designing effective agentic systems requires the seamless composition and integration of agents, tools, and models within dynamic and uncertain environments. Most existing methods rely on static, semantic retrieval approaches for tool or agent discovery. However, effective reuse and composition of existing components remain challenging due to incomplete capability descriptions and the limitations of retrieval methods. Component selection suffers because the decisions are not based on capability, cost, and real-time utility. To address these challenges, we introduce a structured, automated framework for agentic system composition that is inspired by the knapsack problem. Our framework enables a composer agent to systematically identify, select, and assemble an optimal set of agentic components by jointly considering performance, budget constraints, and compatibility. By dynamically testing candidate components and modeling their utility in real-time, our approach streamlines the assembly of agentic systems and facilitates scalable reuse of resources. Empirical evaluation with Claude 3.5 Sonnet across five benchmarking datasets shows that our online-knapsack-based composer consistently lies on the Pareto frontier, achieving higher success rates at significantly lower component costs compared to our baselines. In the single-agent setup, the online knapsack composer shows a success rate improvement of up to 31.6% in comparison to the retrieval baselines. In multi-agent systems, the online knapsack composer increases success rate from 37% to 87% when agents are selected from an agent inventory of 100+ agents. The substantial performance gap confirms the robust adaptability of our method across diverse domains and budget constraints.

cs.CL↗

arXiv2025

LLM-Empowered Event-Chain Driven Code Generation for ADAS in SDV systems

This paper presents an event-chain-driven, LLM-empowered workflow for generating validated, automotive code from natural-language requirements. A Retrieval-Augmented Generation (RAG) layer retrieves relevant signals from large and evolving Vehicle Signal Specification (VSS) catalogs as code generation prompt context, reducing hallucinations and ensuring architectural correctness. Retrieved signals are mapped and validated before being transformed into event chains that encode causal and timing constraints. These event chains guide and constrain LLM-based code synthesis, ensuring behavioral consistency and real-time feasibility. Based on our initial findings from the emergency braking case study, with the proposed approach, we managed to achieve valid signal usage and consistent code generation without LLM retraining.

cs.SE↗

arXiv2025

Multi-Modal Data Exploration via Language Agents

International enterprises, organizations, and hospitals collect large amounts of multi-modal data stored in databases, text documents, images, and videos. While there has been recent progress in the separate fields of multi-modal data exploration as well as in database systems that automatically translate natural language questions to database query languages, the research challenge of querying both structured databases and unstructured modalities (e.g., texts, images) in natural language remains largely unexplored. In this paper, we propose M$^2$EX -a system that enables multi-modal data exploration via language agents. Our approach is based on the following research contributions: (1) Our system is inspired by a real-world use case that enables users to explore multi-modal information systems. (2) M$^2$EX leverages an LLM-based agentic AI framework to decompose a natural language question into subtasks such as text-to-SQL generation and image analysis and to orchestrate modality-specific experts in an efficient query plan. (3) Experimental results on multi-modal datasets, encompassing relational data, text, and images, demonstrate that our system outperforms state-of-the-art multi-modal exploration systems, excelling in both accuracy and various performance metrics, including query latency, API costs, and planning efficiency, thanks to the more effective utilization of the reasoning capabilities of LLMs.

cs.AI↗

arXiv2025

Periodic extreme rainfall in a warmer climate due to stronger convectively-coupled waves

Tropical regions may experience periodic extreme precipitation and suffer from associated periodic deluges in a warmer climate. Recent studies conducted small-domain (around 100 km x 100 km) atmospheric model simulations and found that precipitation transitions from a steady state to a periodic oscillation state in a hothouse climate when the sea surface temperature reaches 320-325 K. Here we conduct global-scale atmospheric model simulations with different complexity, and we find that tropical precipitation in convective regions already transitions to a O(10 day) periodic oscillation state with a O(100 mm/day) amplitude at 305-310 K. This temperature is substantially lower than previously reported, and within reach in a century under a high carbon emission scenario. We attribute the onset of the periodic extreme precipitation to the intensification of convectively-coupled waves, which occurs at temperatures much lower than the radiative mechanism responsible for the transition around 320-325 K identified before.

physics.ao-ph↗

arXiv2025

Granular Computing-driven SAM: From Coarse-to-Fine Guidance for Prompt-Free Segmentation

Prompt-free image segmentation aims to generate accurate masks without manual guidance. Typical pre-trained models, notably Segmentation Anything Model (SAM), generate prompts directly at a single granularity level. However, this approach has two limitations: (1) Localizability, lacking mechanisms for autonomous region localization; (2) Scalability, limited fine-grained modeling at high resolution. To address these challenges, we introduce Granular Computing-driven SAM (Grc-SAM), a coarse-to-fine framework motivated by Granular Computing (GrC). First, the coarse stage adaptively extracts high-response regions from features to achieve precise foreground localization and reduce reliance on external prompts. Second, the fine stage applies finer patch partitioning with sparse local swin-style attention to enhance detail modeling and enable high-resolution segmentation. Third, refined masks are encoded as latent prompt embeddings for the SAM decoder, replacing handcrafted prompts with an automated reasoning process. By integrating multi-granularity attention, Grc-SAM bridges granular computing with vision transformers. Extensive experimental results demonstrate Grc-SAM outperforms baseline methods in both accuracy and scalability. It offers a unique granular computational perspective for prompt-free segmentation.

cs.CV↗

arXiv2025

Asymptotic Performance Analysis of Fluid Antenna Systems: An Extreme Value Theory Perspective

Fluid antenna systems (FAS) allow dynamic reconfiguration to achieve superior diversity gains and reliability. To quantify the performance scaling of FAS with a large number of antenna ports, this paper leverages extreme value theory (EVT) to conduct an asymptotic analysis of the outage probability (OP) and ergodic capacity (EC). The analysis reveals that the OP decays approximately exponentially with the number of antenna ports. Moreover, we establish upper and lower bounds for the asymptotic EC, uncovering its double-logarithmic scaling law. Furthermore, we re-substantiate these scaling laws by exploiting the fact that the mode of the Gumbel distribution scales logarithmically. Besides, we theoretically prove that spatial correlation among antenna ports degrades both OP and EC. All analytical findings are conclusively validated by numerical results.

cs.IT↗