arXiv subjects

Chenlong Zhang

Publications and source records attributed to Chenlong Zhang.

11 recordsLinked to original sources

arXiv2026

The Verification Horizon: No Silver Bullet for Coding Agent Rewards

A classical intuition holds that verifying a solution is easier than producing one. For today's coding agents, this intuition is being inverted: as foundation models develop stronger reasoning capabilities and engineering harnesses grow more sophisticated, generating complex candidate solutions is no longer difficult -- reliably verifying them has become the harder problem. Every verifier we can build is only a proxy for human intent, never the intent itself. This makes verification subject to a twofold difficulty: first, intent is underspecified by nature, making it inherently hard to faithfully check whether it has been fulfilled; second, during model training, optimization widens the gap between proxy and intent -- manifesting as reward hacking or signal saturation. To address this, we characterize the quality of verification signals along three dimensions -- scalability, faithfulness, and robustness -- and argue that achieving all three simultaneously is the central challenge. We further study four reward constructions: a test verifier for general coding tasks, a rubric verifier for frontend tasks, the user as verifier for real-world agent tasks, and an automated agent verifier for long-horizon tasks. Across different task types and policy capability levels, we conduct in-depth analysis and experiments on the core challenges of reward design and how to more effectively leverage reward signals. Experiments show that targeted verification design can effectively suppress reward hacking, improve task completion quality, and achieve significant gains across multiple internal and public benchmarks. These experiences collectively point to a core observation: no fixed reward function can remain effective as policy capability continues to grow; and verification must co-evolve with the generator.

cs.AI↗

arXiv2026

Replay What Matters: Off-Policy Replay for Efficient LLM Reinforcement Unlearning

LLM unlearning has emerged as a cost-effective alternative to full retraining for removing hazardous knowledge from pretrained models while preserving general utility. Recent RL-based methods such as RULE reformulate unlearning as learning a refusal behavior, but their on-policy optimization repeatedly samples from the same forget and retain/boundary prompts throughout training. We identify a critical inefficiency in this process: easy cases quickly converge and provide little useful gradient signal, while hard cases near the forget/retain boundary continue to produce low-reward rollouts that are discarded after a single use. To address this issue, we propose ReRULE, an off-policy replay enhancement for reinforcement unlearning. ReRULE stores low-reward hard-case rollout groups in a replay buffer during early GRPO training and reuses them in later stages through importance-sampled off-policy updates, redirecting computation toward boundary cases that still require learning. Theoretically, we show that ReRULE yields a tighter hard-case convergence bound than pure on-policy RULE. Empirically, ReRULE improves MUSE-Books Retain Quality from 46.3 to 56.2 while adding only 5--11% training time across benchmarks. Its limited improvement on the simpler TOFU setting further supports the intended conditional behavior: replay is most beneficial when the hard/easy disparity is pronounced.

cs.CL↗

arXiv2026

SemEval-2026 Task 12: Abductive Event Reasoning: Towards Real-World Event Causal Inference for Large Language Models

Understanding why real-world events occur is important for both natural language processing and practical decision-making, yet direct-cause inference remains underexplored in evidence-rich settings. To address this gap, we organized SemEval-2026 Task 12: Abductive Event Reasoning (AER).\footnote{The task data is available at https://github.com/sooo66/semeval2026-task12-dataset.git} The task asks systems to identify the most plausible direct cause of a target event from supporting evidence. We formulate AER as an evidence-grounded multiple-choice benchmark that captures key challenges of real-world causal reasoning, including distributed evidence, indirect background factors, and semantically related but non-causal distractors. The shared task attracted 122 participants and received 518 submissions. This paper presents the task formulation, dataset construction pipeline, evaluation setup, and system results. AER provides a focused benchmark for abductive reasoning over real-world events and highlights challenges for future work on causal reasoning and multi-document understanding.

cs.CL↗

arXiv2026

MMR-Life: Piecing Together Real-life Scenes for Multimodal Multi-image Reasoning

Recent progress in the reasoning capabilities of multimodal large language models (MLLMs) has empowered them to address more complex tasks such as scientific analysis and mathematical reasoning. Despite their promise, MLLMs' reasoning abilities across different scenarios in real life remain largely unexplored and lack standardized benchmarks for evaluation. To address this gap, we introduce MMR-Life, a comprehensive benchmark designed to evaluate the diverse multimodal multi-image reasoning capabilities of MLLMs across real-life scenarios. MMR-Life consists of 2,646 multiple-choice questions based on 19,108 images primarily sourced from real-world contexts, comprehensively covering seven reasoning types: abductive, analogical, causal, deductive, inductive, spatial, and temporal. Unlike existing reasoning benchmarks, MMR-Life does not rely on domain-specific expertise but instead requires models to integrate information across multiple images and apply diverse reasoning abilities. The evaluation of 37 advanced models highlights the substantial challenge posed by MMR-Life. Even top models like GPT-5 achieve only 58% accuracy and display considerable variance in performance across reasoning types. Moreover, we analyze the reasoning paradigms of existing MLLMs, exploring how factors such as thinking length, reasoning method, and reasoning type affect their performance. In summary, MMR-Life establishes a comprehensive foundation for evaluating, analyzing, and improving the next generation of multimodal reasoning systems.

cs.CL↗

arXiv2025

Dual-Perspective Disentangled Multi-Intent Alignment for Enhanced Collaborative Filtering

Personalized recommendation requires capturing the complex latent intents underlying user-item interactions. Existing structural models, however, often fail to preserve perspective-dependent interaction semantics and provide only indirect supervision for aligning user and item intents, lacking explicit interaction-level constraints. This entangles heterogeneous interaction signals, leading to semantic ambiguity, reduced robustness under sparse interactions, and limited interpretability. To address these issues, we propose DMICF, a Dual-Perspective Disentangled Multi-Intent framework for collaborative filtering. DMICF models interactions from complementary user- and item-centric perspectives and employs a macro-micro prototype-aware variational encoder to disentangle fine-grained latent intents. Interaction-level supervision enforces dimension-wise alignment between user and item intents, grounding latent factors and enabling their collaborative emergence. Importantly, each component is architecturally flexible, and performance is robust to specific module instantiations. We offer a theoretical analysis to help explain how prototype-aware conditioning may alleviate posterior collapse, while the reconstruction objective promotes intent-wise contrastive alignment between positive and negative interactions. Extensive experiments on multiple benchmarks demonstrate consistent improvements over strong baselines, with ablations validating each core component.

cs.IR↗

arXiv2025

RULE: Reinforcement UnLEarning Achieves Forget-Retain Pareto Optimality

The widespread deployment of Large Language Models (LLMs) trained on massive, uncurated corpora has raised growing concerns about the inclusion of sensitive, copyrighted, or illegal content. This has led to increasing interest in LLM unlearning: the task of selectively removing specific information from a model without retraining from scratch or degrading overall utility. However, existing methods often rely on large-scale forget and retain datasets, and suffer from unnatural responses, poor generalization, or catastrophic utility loss. In this work, we propose Reinforcement UnLearning (RULE), an efficient framework that formulates unlearning as a refusal boundary optimization problem. RULE is trained with a small portion of the forget set and synthesized boundary queries, using a verifiable reward function that encourages safe refusal on forget--related queries while preserving helpful responses on permissible inputs. We provide both theoretical and empirical evidence demonstrating the effectiveness of RULE in achieving targeted unlearning without compromising model utility. Experimental results show that, with only $12%$ forget set and $8%$ synthesized boundary data, RULE outperforms existing baselines by up to $17.5%$ forget quality and $16.3%$ naturalness response while maintaining general utility, achieving forget--retain Pareto optimality. Remarkably, we further observe that RULE improves the naturalness of model outputs, enhances training efficiency, and exhibits strong generalization ability, generalizing refusal behavior to semantically related but unseen queries.

cs.CL↗

arXiv2025

Milmer: a Framework for Multiple Instance Learning based Multimodal Emotion Recognition

Emotions play a crucial role in human behavior and decision-making, making emotion recognition a key area of interest in human-computer interaction (HCI). This study addresses the challenges of emotion recognition by integrating facial expression analysis with electroencephalogram (EEG) signals, introducing a novel multimodal framework-Milmer. The proposed framework employs a transformer-based fusion approach to effectively integrate visual and physiological modalities. It consists of an EEG preprocessing module, a facial feature extraction and balancing module, and a cross-modal fusion module. To enhance visual feature extraction, we fine-tune a pre-trained Swin Transformer on emotion-related datasets. Additionally, a cross-attention mechanism is introduced to balance token representation across modalities, ensuring effective feature integration. A key innovation of this work is the adoption of a multiple instance learning (MIL) approach, which extracts meaningful information from multiple facial expression images over time, capturing critical temporal dynamics often overlooked in previous studies. Extensive experiments conducted on the DEAP dataset demonstrate the superiority of the proposed framework, achieving a classification accuracy of 96.72% in the four-class emotion recognition task. Ablation studies further validate the contributions of each module, highlighting the significance of advanced feature extraction and fusion strategies in enhancing emotion recognition performance. Our code are available at https://github.com/liangyubuaa/Milmer.

cs.CV↗

arXiv2024

DTELS: Towards Dynamic Granularity of Timeline Summarization

The rapid proliferation of online news has posed significant challenges in tracking the continuous development of news topics. Traditional timeline summarization constructs a chronological summary of the events but often lacks the flexibility to meet the diverse granularity needs. To overcome this limitation, we introduce a new paradigm, Dynamic-granularity TimELine Summarization, (DTELS), which aims to construct adaptive timelines based on user instructions or requirements. This paper establishes a comprehensive benchmark for DTLES that includes: (1) an evaluation framework grounded in journalistic standards to assess the timeline quality across four dimensions: Informativeness, Granular Consistency, Factuality, and Coherence; (2) a large-scale, multi-source dataset with multiple granularity timeline annotations based on a consensus process to facilitate authority; (3) extensive experiments and analysis with two proposed solutions based on Large Language Models (LLMs) and existing state-of-the-art TLS methods. The experimental results demonstrate the effectiveness of LLM-based solutions. However, even the most advanced LLMs struggle to consistently generate timelines that are both informative and granularly consistent, highlighting the challenges of the DTELS task.

cs.CL↗

arXiv2024

Continual Few-shot Event Detection via Hierarchical Augmentation Networks

Traditional continual event detection relies on abundant labeled data for training, which is often impractical to obtain in real-world applications. In this paper, we introduce continual few-shot event detection (CFED), a more commonly encountered scenario when a substantial number of labeled samples are not accessible. The CFED task is challenging as it involves memorizing previous event types and learning new event types with few-shot samples. To mitigate these challenges, we propose a memory-based framework: Hierarchical Augmentation Networks (HANet). To memorize previous event types with limited memory, we incorporate prototypical augmentation into the memory set. For the issue of learning new event types in few-shot scenarios, we propose a contrastive augmentation module for token representations. Despite comparing with previous state-of-the-art methods, we also conduct comparisons with ChatGPT. Experiment results demonstrate that our method significantly outperforms all of these methods in multiple continual few-shot event detection tasks.

cs.CL↗

arXiv2022

Robust Coordinated Longitudinal Control of MAV Based on Energy State

Fixed-wing Miniature Air Vehicle (MAV) is not only coupled with longitudinal motion, but also more susceptible to wind disturbance due to its lighter weight, which brings more challenges to its altitude and airspeed controller design. Therefore, in this paper, an improved longitudinal control strategy based on energy state, is proposed to address the above-mentioned issues. The control strategy utilizes the Linear Extended State Observer (LESO) to observe the energy states and the disturbance of the MAV, and then designs a Multiple-Input Multiple-Output (MIMO) controller based on a more coordinated Total Energy Control (TEC) strategy to control the airspeed and altitude of the MAV. The performance of this control strategy has been successfully verified in a Model-in-the-Loop (MIL) simulation with Simulink, and a comparative test with the classical TEC algorithm is carried out.

eess.SY↗

arXiv2015

Dynamical Evolution of Anisotropic Response in Black Phosphorus under Ultrafast Photoexcitation

Black phosphorus has recently emerged as a promising material for high performance electronic and optoelectronic device for its high mobility, tunable mid-infrared bandgap and anisotropic electronic properties. Dynamical evolution of photo excited carriers and its induced change of transient electronic properties are critical for materials' high field performance, but remains to be explored for black phosphorus. In this work, we perform angle resolved transient reflection spectroscopy to study the dynamical evolution of anisotropic properties of black phosphorus under photo excitation. We find that the anisotropy of reflectivity is enhanced in the pump induced quasi-equilibrium state, suggesting an extraordinary enhancement of the anisotropy in dynamical conductivity in hot carrier dominated regime. These results raise enormous possibilities of creating high field, angle sensitive electronic, optoelectronic and remote sensing devices exploiting the dynamical electronic anisotropic with black phosphorus.

cond-mat.mtrl-sci↗