arXiv subjects

Zhenyu Shi

Publications and source records attributed to Zhenyu Shi.

14 recordsLinked to original sources

arXiv2026

AppEval: A Unified Benchmark for LLM-Based Mobile Application Repair in ArkTS, Swift, and Kotlin

Repository-level LLM agents are typically evaluated on projects whose tests run on the build host. It remains unclear whether their repairs survive the mobile build-install-launch-test boundary, where a missing SDK, offline device, or pre-assertion crash can be mistaken for a program failure. We present AppEval, a benchmark and native-toolchain evaluation framework for mobile application repair across HarmonyOS/ArkTS, iOS/Swift, and Android/Kotlin. Each task separates a hidden behavior test from the reference production fix and is accepted only when the same installed-app target reaches an assertion failure on the defective revision and passes after the fix; infrastructure failures remain a distinct outcome. A common schema maps this contract to each platform's build system, runtime, and test runner. The audited Android partition contains 200 accepted instrumentation tasks from 24 independently buildable repositories. On these tasks, five agents achieve Pass@1 between 22.00% and 90.50%, a 68.50-percentage-point spread under the same dynamic oracle. These results show that mobile repair performance depends strongly on the evaluated agent while demonstrating why runtime-aware acceptance is necessary for meaningful comparison. The quantitative findings in this paper are Android-specific; audited iOS and HarmonyOS results are required before drawing cross-platform generalization conclusions.

cs.SE↗

arXiv2026

OdinEval: A Reproducible Benchmark for LLM-Based Program Repair in the Odin Programming Language

Repository-level repair benchmarks still center on a few mainstream languages, leaving systems languages such as Odin largely untested. We present OdinEval, a reproducible benchmark built from documented defects in public Odin repositories. Each instance binds an issue to base and fix commits, a gold patch, an issue-specific regression test, a historical toolchain, and execution records. Admission requires the test to fail on the base revision and pass after the gold fix. When no usable developer test exists, a black-box test is reviewed independently by three instances of the same model, executed in both historical states, and revised from recorded feedback under a versioned Test Writing Skill. We evaluate six language models on 168 filtered instances under one shared protocol. Kimi-K3 records the highest Resolved score at 66.7%, while Qwen3.8-Max has the highest Repro score at 96.4%. The release includes frozen data, source archives, containers, validators, model patches, and audit manifests.

cs.SE↗

arXiv2026

CoMem: Context Management with A Decoupled Long-Context Model

Context management enables agentic models to solve long-horizon tasks through iterative summarization of previous interaction histories. However, this process typically incurs substantial decoding overhead for the extra summarization tokens, which significantly affect the end-to-end response latency at deployment. In this paper, we introduce CoMem, a novel framework that decouples memory management from the primary agent workflow, enabling these processes to execute in parallel. We propose a $k$-step-off asynchronous pipeline that overlaps the memory model's summarization with the agent's inference, effectively masking the latency of context processing. To ensure robustness under this asynchronous setting, we introduce a reward-driven training strategy that aligns the memory model to capture sufficient statistics for the agent's decision-making. Theoretical analysis confirms that CoMem offers a superior efficiency-effectiveness trade-off compared to coupled architectures. Our extensive experimental results on SWE-Bench-Verified show that CoMem provides 1.4x latency improvements upon vanilla long-context solutions while preserving most of the performance. Furthermore, we demonstrate that these latency gains scale favorably with increased system throughput, offering a modular path forward for the independent optimization of agent reasoning and memory compression.

cs.LG↗

arXiv2026

Learning with Rare Success but Rich Feedback via Reflection-Enhanced Self-Distillation

Enabling Large Language Models (LLMs) to continuously improve from environmental interactions is a central challenge in post-training. While on-policy self-distillation offers a promising paradigm, existing methods predominantly treat environmental feedback as a passive conditioning signal. Consequently, they heavily rely on successful demonstrations and struggle to learn in rare-success regimes. To bridge this gap, we introduce Reflection-Enhanced Self-Distillation (RESD), a framework that transforms raw failure feedback into an active source of corrective supervision. Instead of passively appending feedback, RESD interprets failed trajectories by generating retrospective reflections to diagnose local errors, and curates a persistent global playbook to preserve reusable lessons across training steps. The enriched context enables the self-teacher to provide actionable token-level supervision even in the absence of successful rollouts. Empirical evaluations on multiple continual learning tasks demonstrate that RESD substantially outperforms standard self-distillation baselines. Furthermore, RESD achieves significantly faster early-stage improvement than GRPO with $8\times$ samples using only a single rollout per prompt, highlighting its superior interaction efficiency.

cs.LG↗

arXiv2026

T$^2$PO: Uncertainty-Guided Exploration Control for Stable Multi-Turn Agentic Reinforcement Learning

Recent progress in multi-turn reinforcement learning (RL) has significantly improved reasoning LLMs' performances on complex interactive tasks. Despite advances in stabilization techniques such as fine-grained credit assignment and trajectory filtering, instability remains pervasive and often leads to training collapse. We argue that this instability stems from inefficient exploration in multi-turn settings, where policies continue to generate low-information actions that neither reduce uncertainty nor advance task progress. To address this issue, we propose Token- and Turn-level Policy Optimization (T$^2$PO), an uncertainty-aware framework that explicitly controls exploration at fine-grained levels. At the token level, T$^2$PO monitors uncertainty dynamics and triggers a thinking intervention once the marginal uncertainty change falls below a threshold. At the turn level, T$^2$PO identifies interactions with negligible exploration progress and dynamically resamples such turns to avoid wasted rollouts. We evaluate T$^2$PO in diverse environments, including WebShop, ALFWorld, and Search QA, demonstrating substantial gains in training stability and performance improvements with better exploration efficiency. Code is available at: https://github.com/WillDreamer/T2PO.

cs.AI↗

arXiv2026

Alternating Reinforcement Learning with Contextual Rubric Rewards: Beyond the Scalarization Strategy

Reinforcement Learning with Rubric Rewards (RLRR) is a framework that extends conventional reinforcement learning from human feedback (RLHF) and verifiable rewards (RLVR) by replacing scalar preference signals with structured, multi-dimensional, contextual rubric-based evaluations. However, existing approaches in RLRR are limited to linearly compressing vector rewards into a scalar reward with a fixed weightings, which is sensitive to artificial score design and fails to capture correlations among reward dimensions. To overcome the limitations of reward aggregation, this work proposes Alternating Reinforcement Learning with Rubric Rewards (ARL-RR), a framework that eliminates the need for a fixed scalarization by optimizing one semantic rubric meta-class at a time. Theoretically, we show that reward aggregation induces a variance contraction effect, which helps explain the performance gains. We further introduce a lightweight, search-based adaptation procedure that selects the next meta-class dynamically based on task performance, enabling the policy to emphasize critical objectives and thereby improve the model performance. Empirically, our experiments on the HealthBench dataset with experts annotations demonstrate that ARL-RR uniformly outperforms scalarized methods in both model performance and training efficiency across different model scales (1.7B, 4B, 8B, and 14B).

cs.LG↗

arXiv2025

Efficient Retail Video Annotation: A Robust Key Frame Generation Approach for Product and Customer Interaction Analysis

Accurate video annotation plays a vital role in modern retail applications, including customer behavior analysis, product interaction detection, and in-store activity recognition. However, conventional annotation methods heavily rely on time-consuming manual labeling by human annotators, introducing non-robust frame selection and increasing operational costs. To address these challenges in the retail domain, we propose a deep learning-based approach that automates key-frame identification in retail videos and provides automatic annotations of products and customers. Our method leverages deep neural networks to learn discriminative features by embedding video frames and incorporating object detection-based techniques tailored for retail environments. Experimental results showcase the superiority of our approach over traditional methods, achieving accuracy comparable to human annotator labeling while enhancing the overall efficiency of retail video annotation. Remarkably, our approach leads to an average of 2 times cost savings in video annotation. By allowing human annotators to verify/adjust less than 5% of detected frames in the video dataset, while automating the annotation process for the remaining frames without reducing annotation quality, retailers can significantly reduce operational costs. The automation of key-frame detection enables substantial time and effort savings in retail video labeling tasks, proving highly valuable for diverse retail applications such as shopper journey analysis, product interaction detection, and in-store security monitoring.

cs.CV↗

arXiv2025

DA-STGCN: 4D Trajectory Prediction Based on Spatiotemporal Feature Extraction

The importance of four-dimensional (4D) trajectory prediction within air traffic management systems is on the rise. Key operations such as conflict detection and resolution, aircraft anomaly monitoring, and the management of congested flight paths are increasingly reliant on this foundational technology, underscoring the urgent demand for intelligent solutions. The dynamics in airport terminal zones and crowded airspaces are intricate and ever-changing; however, current methodologies do not sufficiently account for the interactions among aircraft. To tackle these challenges, we propose DA-STGCN, an innovative spatiotemporal graph convolutional network that integrates a dual attention mechanism. Our model reconstructs the adjacency matrix through a self-attention approach, enhancing the capture of node correlations, and employs graph attention to distill spatiotemporal characteristics, thereby generating a probabilistic distribution of predicted trajectories. This novel adjacency matrix, reconstructed with the self-attention mechanism, is dynamically optimized throughout the network's training process, offering a more nuanced reflection of the inter-node relationships compared to traditional algorithms. The performance of the model is validated on two ADS-B datasets, one near the airport terminal area and the other in dense airspace. Experimental results demonstrate a notable improvement over current 4D trajectory prediction methods, achieving a 20% and 30% reduction in the Average Displacement Error (ADE) and Final Displacement Error (FDE), respectively. The incorporation of a Dual-Attention module has been shown to significantly enhance the extraction of node correlations, as verified by ablation experiments.

cs.CV↗

arXiv2024

Radar and Camera Fusion for Object Detection and Tracking: A Comprehensive Survey

Multi-modal fusion is imperative to the implementation of reliable object detection and tracking in complex environments. Exploiting the synergy of heterogeneous modal information endows perception systems the ability to achieve more comprehensive, robust, and accurate performance. As a nucleus concern in wireless-vision collaboration, radar-camera fusion has prompted prospective research directions owing to its extensive applicability, complementarity, and compatibility. Nonetheless, there still lacks a systematic survey specifically focusing on deep fusion of radar and camera for object detection and tracking. To fill this void, we embark on an endeavor to comprehensively review radar-camera fusion in a holistic way. First, we elaborate on the fundamental principles, methodologies, and applications of radar-camera fusion perception. Next, we delve into the key techniques concerning sensor calibration, modal representation, data alignment, and fusion operation. Furthermore, we provide a detailed taxonomy covering the research topics related to object detection and tracking in the context of radar and camera technologies.Finally, we discuss the emerging perspectives in the field of radar-camera fusion perception and highlight the potential areas for future research.

cs.CV↗

arXiv2023

Efficient Link Prediction in Continuous-Time Dynamic Networks using Optimal Transmission and Metropolis Hastings Sampling

Efficient link prediction in continuous-time dynamic networks is a challenging problem that has attracted much research attention in recent years. A widely used approach to dynamic network link prediction is to extract the local structure of the target link through temporal random walk on the network and learn node features using a coding model. However, this approach often assumes that candidate temporal neighbors follow some certain types of distributions, which may be inappropriate for real-world networks, thereby incurring information loss. To address this limitation, we propose a framework in continuous-time dynamic networks based on Optimal Transmission (OT) and Metropolis Hastings (MH) sampling (COM). Specifically, we use optimal transmission theory to calculate the Wasserstein distance between the current node and the time-valid candidate neighbors to minimize information loss in node information propagation. Additionally, we employ the MH algorithm to obtain higher-order structural relationships in the vicinity of the target link, as it is a Markov Chain Monte Carlo method and can flexibly simulate target distributions with complex patterns. We demonstrate the effectiveness of our proposed method through experiments on eight datasets from different fields.

math.DS↗

arXiv2023

APAM: Adaptive Pre-training and Adaptive Meta Learning in Language Model for Noisy Labels and Long-tailed Learning

Practical natural language processing (NLP) tasks are commonly long-tailed with noisy labels. Those problems challenge the generalization and robustness of complex models such as Deep Neural Networks (DNNs). Some commonly used resampling techniques, such as oversampling or undersampling, could easily lead to overfitting. It is growing popular to learn the data weights leveraging a small amount of metadata. Besides, recent studies have shown the advantages of self-supervised pre-training, particularly to the under-represented data. In this work, we propose a general framework to handle the problem of both long-tail and noisy labels. The model is adapted to the domain of problems in a contrastive learning manner. The re-weighting module is a feed-forward network that learns explicit weighting functions and adapts weights according to metadata. The framework further adapts weights of terms in the loss function through a combination of the polynomial expansion of cross-entropy loss and focal loss. Our extensive experiments show that the proposed framework consistently outperforms baseline methods. Lastly, our sensitive analysis emphasizes the capability of the proposed framework to handle the long-tailed problem and mitigate the negative impact of noisy labels.

cs.CL↗

arXiv2021

Effects of Dynamic-Win-Stay-Lose-Learn model with voluntary participation in social dilemma

In recent years, Win-Stay-Lose-Learn rule has attracted wide attention as an effective strategy updating rule, and voluntary participation is proposed by introducing a third strategy in Prisoner's dilemma game. Some researches show that combining Win-Stay-Lose-Learn rule with voluntary participation could promote cooperation more significantly under moderate temptation values, however, cooperators' survival under high aspiration levels and high temptation values is still a challenging problem. In this paper, inspired by Achievement Motivation Theory, a Dynamic-Win-Stay-Lose-Learn rule with voluntary participation is investigated, where a dynamic aspiration process is introduced to describe the co-evolution of individuals' strategies and aspirations. It is found that cooperation is extremely promoted and defection is almost extinct in our model, even when the initial aspiration levels and temptation values are high. The combination of dynamic aspiration and voluntary participation plays an active role since loners could survive under high initial aspiration levels and they will expand stably because of their fixed payoffs. The robustness of our model is also discussed and some adverse structures are found which should be alerted in the evolutionary process. Our work provides a more realistic model and shows that cooperators may prevail defectors in an unfavorable initial environment.

cs.GT↗

arXiv2020

Dynamic aspiration based on Win-Stay-Lose-Learn rule in Spatial Prisoner's Dilemma Gam

Prisoner's dilemma game is the most commonly used model of spatial evolutionary game which is considered as a paradigm to portray competition among selfish individuals. In recent years, Win-Stay-Lose-Learn, a strategy updating rule base on aspiration, has been proved to be an effective model to promote cooperation in spatial prisoner's dilemma game, which leads aspiration to receive lots of attention. But in many research the assumption that individual's aspiration is fixed is inconsistent with recent results from psychology. In this paper, according to Expected Value Theory and Achievement Motivation Theory, we propose a dynamic aspiration model based on Win-Stay-Lose-Learn rule in which individual's aspiration is inspired by its payoff. It is found that dynamic aspiration has a significant impact on the evolution process, and different initial aspirations lead to different results, which are called Stable Coexistence under Low Aspiration, Dependent Coexistence under Moderate aspiration and Defection Explosion under High Aspiration respectively. Furthermore, a deep analysis is performed on the local structures which cause cooperator's existence or defector's expansion, and the evolution process for different parameters including strategy and aspiration. As a result, the intrinsic structures leading to defectors' expansion and cooperators' survival are achieved for different evolution process, which provides a penetrating understanding of the evolution. Compared to fixed aspiration model, dynamic aspiration introduces a more satisfactory explanation on population evolution laws and can promote deeper comprehension for the principle of prisoner's dilemma.

physics.soc-ph↗

arXiv2019

Inducing Cooperation via Team Regret Minimization based Multi-Agent Deep Reinforcement Learning

Existing value-factorized based Multi-Agent deep Reinforce-ment Learning (MARL) approaches are well-performing invarious multi-agent cooperative environment under thecen-tralized training and decentralized execution(CTDE) scheme,where all agents are trained together by the centralized valuenetwork and each agent execute its policy independently. How-ever, an issue remains open: in the centralized training process,when the environment for the team is partially observable ornon-stationary, i.e., the observation and action informationof all the agents cannot represent the global states, existingmethods perform poorly and sample inefficiently. Regret Min-imization (RM) can be a promising approach as it performswell in partially observable and fully competitive settings.However, it tends to model others as opponents and thus can-not work well under the CTDE scheme. In this work, wepropose a novel team RM based Bayesian MARL with threekey contributions: (a) we design a novel RM method to traincooperative agents as a team and obtain a team regret-basedpolicy for that team; (b) we introduce a novel method to de-compose the team regret to generate the policy for each agentfor decentralized execution; (c) to further improve the perfor-mance, we leverage a differential particle filter (a SequentialMonte Carlo method) network to get an accurate estimation ofthe state for each agent. Experimental results on two-step ma-trix games (cooperative game) and battle games (large-scalemixed cooperative-competitive games) demonstrate that ouralgorithm significantly outperforms state-of-the-art methods.

cs.AI↗