arXiv subjects

Zhongyi Li

Publications and source records attributed to Zhongyi Li.

10 recordsLinked to original sources

arXiv2026

Conformal Prediction-Enabled Sequence-level Policy Optimization for Reasoning Language Models

Reinforcement learning improves the reasoning capability of large language models through verifiable rewards for mathematical and coding tasks. In token-level methods such as GRPO, clipping acts on individual token ratios while the reward and advantage apply to a complete response. Recent sequence-level methods align these units but still use fixed clipping bands, whose tail probabilities change as the score distribution evolves. We propose Conformal Prediction-Enabled Sequence-level Policy Optimization (CSPO), which adaptively calibrates sequence-level clipping bounds using recent training statistics. The resulting band targets a specified marginal tail budget and adapts to policy drift; calibration conditional on response length requires additional assumptions. We bound the probability that a fresh score falls outside the band by the target miscoverage level plus explicit CDF discrepancy and drift terms. Under additional moment conditions, this bound also controls the objective and gradient error relative to the unclipped GSPO power surrogate. Experiments with the Qwen3-1.7B series show improvements over GSPO on difficult mathematical reasoning benchmarks and ALFWorld, with clipping endpoints that adapt during training.

cs.LG↗

arXiv2026

Beyond Sub-Gaussian Detector Scores: Robust Weighted Profile-Loss Change Point Detection for Human-LLM Text Segmentation

Mixed human-LLM documents require locating authorship transitions from detector scores whose reliability varies across text units. Existing weighted mean contrasts are vulnerable to extreme scores, while directly replacing means with robust centers obscures how a misplaced boundary changes the population objective. We propose Robust Weighted Profile-Loss Change Point Detection (RWCP), which combines capped reliability weights, Huber profile gains, and narrowest-over-threshold search in reliability coordinates. Our key analysis expresses the population gap between a true and a displaced split as a merge cost, avoiding a closed-form solution for the nonlinear center of a mixed segment. Under explicit curvature, spacing, and dependence conditions, core RWCP recovers the number of changes and localizes their boundaries; its quadratic-loss limit recovers squared weighted CUSUM. We also study RWCP-R, a separately evaluated decoder that shares source centers across nonadjacent passages. Across five retrospective cached-score benchmark families, core RWCP reduces family-macro WindowDiff by 17.6\% relative to weighted change-point detection, and RWCP-R lowers it further. Boundary recovery improves most clearly for isolated changes, while both fixed configurations miss changes in collaborative and densely alternating text.

cs.AI↗

arXiv2026

STAR-GRPO: Canonical Anchoring and Reliability-First Advantages against Representation-Dependent Reward Hacking

Reward hacking occurs when policy optimization exploits a brittle reward interface or an overly permissive proxy objective, improving the training score without improving the underlying response quality. This phenomenon is amplified in group-relative policy optimization: an unsupported reward can shift the group baseline and alter the updates of other rollouts, while post-hoc or purely relative weighting cannot represent group-wide uncertainty. We propose \emph{Self-Tuned Anchored Reliability Group-Relative Policy Optimization} (STAR-GRPO), a reliability-first advantage estimator based on paired assessments of the same rollout. STAR separates the quality signal from its learning influence: score disagreement determines rollout reliability, relative reliability enters a self-tuned robust location--scale fit before group normalization, and absolute group reliability attenuates the resulting bounded advantage. The analysis establishes coordinate and second-moment bounds, characterizes exact centering through the weighted location equation, and gives reliability-dependent attenuation guarantees for outlying rewards. We evaluate STAR-GRPO in two complementary reward-hacking regimes. In token-interface exploitation, STAR prevents runaway optimization of the deployed-interface score while improving the canonical quality signal. In rubric-proxy overoptimization for medical reasoning, STAR improves independent semantic evaluation, narrows the proxy--judge discrepancy, and reduces overclaim while optimizing the same task proxy. Together, these results show that reliability-first normalization offers a principled way to limit unsupported reward influence on both group baselines and policy updates, while retaining the task reward as the optimization target.

cs.AI↗

arXiv2026

Dual-Channel Robust Group-Relative Policy Optimization via Advantage and Sequence-Weight Estimation

Group-relative policy optimization relies on reward-derived advantages and sequence-level likelihood weights, both of which can be sensitive to localized outliers. Extreme rewards can collapse the contrast among clean responses after group normalization, while token-level log-ratio perturbations can alter sequence weights and clipping decisions. We introduce RoVR-GSPO, a dual-channel robust optimizer that addresses these failure modes separately. Its reward channel combines robust reference estimation with bounded residual credit, while its ratio channel uses differentiable SoftRoVR aggregation to construct robust sequence weights. We provide stability and efficiency analyses for both channels. Experiments on mathematical reasoning, long-context summarization, and tool-call annotation show consistent improvements over GSPO, while controlled perturbation studies demonstrate stronger robustness to reward contamination and token-ratio anomalies.

cs.AI↗

arXiv2026

Counterfactual Credit Policy Optimization for Multi-Agent Collaboration

Collaborative multi-agent large language models (LLMs) can solve complex reasoning tasks by decomposing roles, but reinforcement learning for such systems is limited by credit assignment: shared terminal rewards obscure individual contributions and can encourage free-riding. We introduce two optimizer-agnostic credit assignment methods for converting joint outcomes into agent-specific learning signals. Counterfactual Credit for Policy Optimization (CCPO) estimates an agent's marginal contribution by comparing the realized joint outcome with a counterfactual outcome where that agent is removed. Self-Evaluated Credit for Policy Optimization (SEPO) uses constrained self- and peer-evaluations as a verifier-anchored credit signal while keeping the external task outcome dominant. Both operate at the reward-construction layer rather than as policy optimizers, producing role-specific rewards or advantages for GRPO, GSPO, or REINFORCE++. We instantiate these credit signals in a sequential Think--Solve setting and evaluate them on mathematical reasoning benchmarks. Results show that explicit credit assignment often improves dual-agent reasoning, especially on MATH500 and several out-of-distribution settings, while gains vary across models and datasets. Our code is available at: https://github.com/bhai114/ccpo.

cs.AI↗

arXiv2026

Sharper Generalization Bounds for Transformer

This paper studies generalization error bounds for Transformer models. Based on the offset Rademacher complexity, we derive sharper generalization bounds for different Transformer architectures, including single-layer single-head, single-layer multi-head, and multi-layer Transformers. We first express the excess risk of Transformers in terms of the offset Rademacher complexity. By exploiting its connection with the empirical covering numbers of the corresponding hypothesis spaces, we obtain excess risk bounds that achieve optimal convergence rates up to constant factors. We then derive refined excess risk bounds by upper bounding the covering numbers of Transformer hypothesis spaces using matrix ranks and matrix norms, leading to precise, architecture-dependent generalization bounds. Finally, we relax the boundedness assumption on feature mappings and extend our theoretical results to settings with unbounded (sub-Gaussian) features and heavy-tailed distributions.

cs.LG↗

arXiv2026

Adaptive Robust Estimator for Multi-Agent Reinforcement Learning

Multi-agent collaboration has emerged as a powerful paradigm for enhancing the reasoning capabilities of large language models, yet it suffers from interaction-level ambiguity that blurs generation, critique, and revision, making credit assignment across agents difficult. Moreover, policy optimization in this setting is vulnerable to heavy-tailed and noisy rewards, which can bias advantage estimation and trigger unstable or even divergent training. To address both issues, we propose a robust multi-agent reinforcement learning framework for collaborative reasoning, consisting of two components: Dual-Agent Answer-Critique-Rewrite (DACR) and an Adaptive Robust Estimator (ARE). DACR decomposes reasoning into a structured three-stage pipeline: answer, critique, and rewrite, while enabling explicit attribution of each agent's marginal contribution to its partner's performance. ARE provides robust estimation of batch experience means during multi-agent policy optimization. Across mathematical reasoning and embodied intelligence benchmarks, even under noisy rewards, our method consistently outperforms the baseline in both homogeneous and heterogeneous settings. These results indicate stronger robustness to reward noise and more stable training dynamics, effectively preventing optimization failures caused by noisy reward signals.

cs.AI↗

arXiv2025

AMSbench: A Comprehensive Benchmark for Evaluating MLLM Capabilities in AMS Circuits

Analog/Mixed-Signal (AMS) circuits play a critical role in the integrated circuit (IC) industry. However, automating Analog/Mixed-Signal (AMS) circuit design has remained a longstanding challenge due to its difficulty and complexity. Although recent advances in Multi-modal Large Language Models (MLLMs) offer promising potential for supporting AMS circuit analysis and design, current research typically evaluates MLLMs on isolated tasks within the domain, lacking a comprehensive benchmark that systematically assesses model capabilities across diverse AMS-related challenges. To address this gap, we introduce AMSbench, a benchmark suite designed to evaluate MLLM performance across critical tasks including circuit schematic perception, circuit analysis, and circuit design. AMSbench comprises approximately 8000 test questions spanning multiple difficulty levels and assesses eight prominent models, encompassing both open-source and proprietary solutions such as Qwen 2.5-VL and Gemini 2.5 Pro. Our evaluation highlights significant limitations in current MLLMs, particularly in complex multi-modal reasoning and sophisticated circuit design tasks. These results underscore the necessity of advancing MLLMs' understanding and effective application of circuit-specific knowledge, thereby narrowing the existing performance gap relative to human expertise and moving toward fully automated AMS circuit design workflows. Our data is released at this URL.

cs.LG↗

arXiv2025

Automated SAR ADC Sizing Using Analytical Equations

Conventional analog and mixed-signal (AMS) circuit designs heavily rely on manual effort, which is time-consuming and labor-intensive. This paper presents a fully automated design methodology for Successive Approximation Register (SAR) Analog-to-Digital Converters (ADCs) from performance specifications to complete transistor sizing. To tackle the high-dimensional sizing problem, we propose a dual optimization scheme. The system-level optimization iteratively partitions the overall requirements and analytically maps them to subcircuit design specifications, while local optimization loops determines the subcircuits' design parameters. The dependency graph-based framework serializes the simulations for verification, knowledge-based calculations, and transistor sizing optimization in topological order, which eliminates the need for human intervention. We demonstrate the effectiveness of the proposed methodology through two case studies with varying performance specifications, achieving high SNDR and low power consumption while meeting all the specified design constraints.

cs.AR↗

arXiv2024

Joint Beamforming for Multi-target Detection and Multi-user Communication in ISAC Systems

Detecting weak targets is one of the main challenges for integrated sensing and communication (ISAC) systems. Sensing and communication suffer from a performance trade-off in ISAC systems. As the communication demand increases, sensing ability, especially weak target detection performance, will inevitably reduce. Traditional approaches fail to address this issue. In this paper, we develop a joint beamforming scheme and formulate it as a max-min problem to maximize the detection probability of the weakest target under the constraint of the signal-to-interference-plus-noise ratio (SINR) of multi-user communication. An alternating optimization (AO) algorithm is developed for solving the complicated non-convex problem to obtain the joint beamformer. The proposed scheme can direct the transmit energy toward the multiple targets properly to ensure robust multi-target detection performance. Numerical results show that the proposed beamforming scheme can effectively increase the detection probability of the weakest target compared to baseline approaches while ensuring communication performance.

eess.SP↗