arXiv subjects

Zhuochun Li

Publications and source records attributed to Zhuochun Li.

16 recordsLinked to original sources

arXiv2026

SPEAR: Distilling Domain-Adaptive Reasoning Skeletons via Sequential Symbolic Alignment in Reinforcement Learning

Reinforcement learning-based knowledge distillation has the potential to transfer complex reasoning from teacher to student models, yet it currently faces a critical dilemma: researchers must choose between sparse outcome-based rewards, which provide insufficient logical guidance, or expensive neural Process Reward Models (PRMs) for dense signals. We resolve this by introducing SPEAR (Symbolic Process Evaluation and Alignment Reward), a training-free and plug-and-play process reward method for sequence-level on-policy distillation. SPEAR projects natural-language reasoning traces into domain-adaptive symbolic milestones, providing an efficient proxy for process-level reasoning alignment. By utilizing the longest common subsequence (LCS) to align student explorations with teacher milestones, SPEAR provides a dense, order-aware reward signal that enforces logical consistency without the need for an external neural verifier. Our experiments across math, science, and commonsense reasoning tasks demonstrate that SPEAR effectively bridges the reasoning gap between student and teacher models via sequence-level distillation with efficient dense process rewards. Our code and data are available at: https://github.com/zhuochunli/SPEAR.

cs.CL↗

arXiv2026

One Success Isn't Reliability: Thinkingbox, a Sandbox and Benchmark for Agents in Stateful Business Workflows

Recent agent benchmarks increasingly ground evaluation in executable environments, from code repair to web navigation, app APIs, and function calling. Yet completing consequential work beyond code requires more than producing a plausible response or valid tool call: agents must gather missing information over multiple turns, follow domain policies, coordinate dependent tools, and realize the correct persistent state transition without collateral effects. In this paper, we introduce Thinkingbox, a sandbox for tool-agent-user interaction that provides isolated MCP-compatible tool sessions, complete execution traces, and outcome evaluation over terminal backend state. Built on this sandbox, Thinkingbox-bench contains 507 policy-conditioned workflows across numerous scenarios, including retail, hospitality, auto insurance, neobank internal IT, and consulting IT/HR support. Each attempt is evaluated by task-specific executable checks that accept valid trajectories while rejecting wrong, missing, or extra effects; designated tasks additionally check required properties of the final response. Across proprietary and open-weight models, the strongest model Claude Opus 5 achieves 66.50% pass@1, but only 47.53% pass^20. Moreover, many failed trials show clean termination and valid state-changing actions, showing that response or tool-call-level signals are not clear proxies for end-to-end task completion. Thinkingbox-bench reveals a large gap between occasionally finding a successful trajectory and reliably completing stateful business tasks. We release both Thinkingbox and Thinkingbox-Bench: https://github.com/microsoft/thinkingbox

cs.CL↗

arXiv2026

Cognitive Episodes in LLM Reasoning Traces Enable Interpretable Human Item Difficulty Prediction

Predicting human item difficulty is central to educational assessment, where reliable estimates support fairness and effective test construction. Existing methods often depend on costly human calibration or item-level textual representations, providing limited evidence about the cognitive processes that make items difficult. We argue that difficulty should be viewed not only as a property of item text, but also as an observable consequence of the problem-solving burden an item induces. Large Reasoning Models (LRMs) offer scalable process evidence through reasoning traces, but such evidence must be structured to support interpretable modeling. To this end, we introduce Epi2Diff (Episode to Difficulty), a framework that maps LRM reasoning traces into cognitively grounded episode sequences. These episodes group trace segments into functional problem-solving states, enabling difficulty to be modeled through reasoning scale, effort allocation, and state transitions. Epi2Diff extracts compact episode-dynamic features and combines them with semantic item representations for human difficulty prediction. Experiments on four real-world human difficulty datasets show that Epi2Diff consistently outperforms strong baselines, including fine-tuned small language models, LLM in-context learning, and supervised LLM adaptation. On SAT-derived classification benchmarks, Epi2Diff achieves an 8.1% average relative gain over supervised LLM fine-tuning baselines. Further analyses show that harder items induce more effortful, iterative, and implementation-centered episode dynamics, rather than merely longer responses. These results demonstrate that cognitive episodes in LRM reasoning traces provide a predictive and interpretable process representation for human item difficulty, offering a new lens for educational measurement with reasoning models.

cs.CL↗

arXiv2026

MoEGen: Mixture-of-Experts for Instance-Adaptive LoRA Generation

Parameter-efficient fine-tuning (PEFT) enables efficient adaptation of large language models, but existing MoE-based PEFT methods typically improve capacity by storing multiple full LoRA experts, causing adapter storage to grow linearly with the number of experts and restricting adaptation to a fixed expert pool. We ask whether MoE-based PEFT can produce instance-specific adaptations without explicitly storing a separate LoRA module for each expert. To address this gap, we propose MoEGen, an adaptation framework that shifts MoE-based PEFT from expert selection to expert-conditioned parameter generation. Instead of storing each expert as a full LoRA adapter, MoEGen represents each expert as a small learnable vector, termed an expert code. It routes each input over these vectors and uses their weighted combination to condition a lightweight hypernetwork that generates input-specific low-rank updates. This design decouples expert capacity from adapter storage while enabling instance-conditioned adaptation. Experiments on eight commonsense reasoning benchmarks show consistent improvements over strong static and MoE-based PEFT baselines across three backbones. MoEGen also performs strongly in joint medical and legal-domain adaptation.

cs.CL↗

arXiv2026

Rethinking LLM-as-a-Judge: Representation-as-a-Judge with Small Language Models via Semantic Capacity Asymmetry

Large language models (LLMs) are widely used as reference-free evaluators via prompting, but this "LLM-as-a-Judge" paradigm is costly, opaque, and sensitive to prompt design. In this work, we investigate whether smaller models can serve as efficient evaluators by leveraging internal representations instead of surface generation. We uncover a consistent empirical pattern: small LMs, despite with weak generative ability, encode rich evaluative signals in their hidden states. This motivates us to propose the Semantic Capacity Asymmetry Hypothesis: evaluation requires significantly less semantic capacity than generation and can be grounded in intermediate representations, suggesting that evaluation does not necessarily need to rely on large-scale generative models but can instead leverage latent features from smaller ones. Our findings motivate a paradigm shift from LLM-as-a-Judge to Representation-as-a-Judge, a decoding-free evaluation strategy that probes internal model structure rather than relying on prompted output. We instantiate this paradigm through INSPECTOR, a probing-based framework that predicts aspect-level evaluation scores from small model representations. Experiments on reasoning benchmarks (GSM8K, MATH, GPQA) show that INSPECTOR substantially outperforms prompting-based small LMs and closely approximates full LLM judges, while offering a more efficient, reliable, and interpretable alternative for scalable evaluation. The code and data are available at: https://github.com/zhuochunli/Representation-as-a-judge

cs.CL↗

arXiv2026

StepGap: A Hybrid NLI-LLM Checker for Step-Level Evidence-Gap Detectionin Multi-Hop Question Answering

We present \textbf{StepGap}, a hybrid NLI-LLM decision tree that detects step-level evidence gaps in multi-hop QA and emits one of three typed labels: \textsc{Contradicted Claim} (CC), \textsc{Irrelevant Evidence} (IE), or \textsc{Missing Bridge} (MB), each tied to a concrete repair action. On 82 multi-hop questions (181 annotated steps, $κ{=}0.704$), StepGap reaches sF1$=$72.0, within the bootstrap confidence interval of an LLM-only baseline (70.1) but with a more decomposable structure: every StepGap stage \emph{hurts} F1 when removed, while three of four LLM-only removals \emph{improve} F1 -- a sign of \emph{competing-error cancellation}, where internal stages mask each other's errors. We further expose a \emph{Q-F1 trap}: question-level F1 is mechanically inflated by checkers that flag every step, making step-level F1 the necessary diagnostic. Used as a typed GRPO process reward, StepGap improves Qwen2.5-7B-Instruct Exact Match from $32.1{\pm}0.3$ to $35.4{\pm}0.9$ across three seeds, with the single-run comparison showing a $+5.6$ Avg EM gain over the matched Search-R1 GRPO reproduction.

cs.CL↗

arXiv2026

TreeDiff: AST-Guided Code Generation with Diffusion LLMs

Code generation is increasingly critical for real-world applications. Still, diffusion-based large language models continue to struggle with this demand. Unlike free-form text, code requires syntactic precision; even minor structural inconsistencies can render a program non-executable. Existing diffusion-based large language models rely on random token masking for corruption, leading to two key failures: they lack awareness of syntactic boundaries during the iterative denoising process, and they fail to capture the long-range hierarchical dependencies essential for program correctness. We propose TreeDiff to address both issues. Specifically, we propose a syntax-aware diffusion framework that incorporates structural priors from Abstract Syntax Tree (AST) into the corruption process. Instead of masking individual tokens at random, we selectively mask tokens belonging to key AST nodes. By aligning the corruption process with the underlying structure of code, our method encourages the model to internalize the compositional nature of programming languages, enabling it to reconstruct programs that respect grammatical boundaries and capture long-range dependencies. Our method achieves a 13.3% relative improvement over the random masking training method, demonstrating its effectiveness in code generation task by leveraging underlying structures.

cs.CL↗

arXiv2026

Retrieval--Reasoning Processes for Multi-hop Question Answering: A Four-Axis Design Framework and Empirical Trends

Multi-hop question answering (QA) requires systems to iteratively retrieve evidence and reason across multiple hops. While recent RAG and agentic methods report strong results, the underlying retrieval--reasoning \emph{process} is often left implicit, making procedural choices hard to compare across model families. This survey takes the execution procedure as the unit of analysis and introduces a four-axis framework covering (A) overall execution plan, (B) index structure, (C) next-step control (strategies and triggers), and (D) stop/continue criteria. Using this schema, we map representative multi-hop QA systems and synthesize reported ablations and tendencies on standard benchmarks (e.g., HotpotQA, 2WikiMultiHopQA, MuSiQue), highlighting recurring trade-offs among effectiveness, efficiency, and evidence faithfulness. We conclude with open challenges for retrieval--reasoning agents, including structure-aware planning, transferable control policies, and robust stopping under distribution shift.

cs.CL↗

arXiv2025

Think Globally, Group Locally: Evaluating LLMs Using Multi-Lingual Word Grouping Games

Large language models (LLMs) can exhibit biases in reasoning capabilities due to linguistic modality, performing better on tasks in one language versus another, even with similar content. Most previous works evaluate this through reasoning tasks where reliance on strategies or knowledge can ensure success, such as in commonsense or math tasks. However, abstract reasoning is vital to reasoning for everyday life, where people apply "out-of-the-box thinking" to identify and use patterns for solutions, without a reliance on formulaic approaches. Comparatively, little work has evaluated linguistic biases in this task type. In this paper, we propose a task inspired by the New York Times Connections: GlobalGroup, that evaluates models in an abstract reasoning task across several languages. We constructed a game benchmark with five linguistic backgrounds -- English, Spanish, Chinese, Hindi, and Arabic -- in both the native language and an English translation for comparison. We also proposed game difficulty measurements to evaluate models on games with similar difficulty, enabling a more controlled comparison, which is particularly important in reasoning evaluations. Through experimentation, we find English modalities largely lead to better performance in this abstract reasoning task, and performance disparities between open- and closed-source models.

cs.CL↗

arXiv2025

Curriculum Guided Reinforcement Learning for Efficient Multi Hop Retrieval Augmented Generation

Retrieval-augmented generation (RAG) grounds large language models (LLMs) in up-to-date external evidence, yet existing multi-hop RAG pipelines still issue redundant subqueries, explore too shallowly, or wander through overly long search chains. We introduce EVO-RAG, a curriculum-guided reinforcement learning framework that evolves a query-rewriting agent from broad early-stage exploration to concise late-stage refinement. EVO-RAG couples a seven-factor, step-level reward vector (covering relevance, redundancy, efficiency, and answer correctness) with a time-varying scheduler that reweights these signals as the episode unfolds. The agent is trained with Direct Preference Optimization over a multi-head reward model, enabling it to learn when to search, backtrack, answer, or refuse. Across four multi-hop QA benchmarks (HotpotQA, 2WikiMultiHopQA, MuSiQue, and Bamboogle), EVO-RAG boosts Exact Match by up to 4.6 points over strong RAG baselines while trimming average retrieval depth by 15 %. Ablation studies confirm the complementary roles of curriculum staging and dynamic reward scheduling. EVO-RAG thus offers a general recipe for building reliable, cost-effective multi-hop RAG systems.

cs.CL↗

arXiv2025

Learning from Committee: Reasoning Distillation from a Mixture of Teachers with Peer-Review

While reasoning capabilities typically emerge in large language models (LLMs) with tens of billions of parameters, recent research focuses on improving smaller open-source models through knowledge distillation (KD) from commercial LLMs. However, many of these studies rely solely on responses from a single LLM as the gold rationale, unlike the natural human learning process, which involves understanding both the correct answers and the reasons behind mistakes. In this paper, we introduce a novel Fault-Aware DistIllation via Peer-Review (FAIR) approach: 1) instead of merely obtaining rationales from teachers, our method asks teachers to identify and explain the student's mistakes, providing customized instruction learning data; 2) we design a simulated peer-review process between teacher LLMs, and selects only the generated rationales above the acceptance threshold, which reduces the chance of teachers guessing correctly with flawed rationale, improving instructional data quality. Comprehensive experiments and analysis on mathematical, commonsense, and logical reasoning tasks demonstrate the effectiveness of our method. Our code is available at https://github.com/zhuochunli/Learn-from-Committee.

cs.CL↗

arXiv2025

ReasoningRank: Teaching Student Models to Rank through Reasoning-Based Knowledge Distillation

Reranking documents based on their relevance to a given query is a critical task in information retrieval. Traditional reranking methods often lack transparency and rely on proprietary models, hindering reproducibility and interpretability. We propose Reason-to-Rank (R2R), a novel open-source reranking approach that enhances transparency by generating two types of reasoning: direct relevance reasoning, which explains how a document addresses the query, and comparison reasoning, which justifies the relevance of one document over another. We leverage large language models (LLMs) as teacher models to generate these explanations and distill this knowledge into smaller, openly available student models. Our student models are trained to generate meaningful reasoning and rerank documents, achieving competitive performance across multiple datasets, including MSMARCO and BRIGHT. Experiments demonstrate that R2R not only improves reranking accuracy but also provides valuable insights into the decision-making process. By offering a structured and interpretable solution with openly accessible resources, R2R aims to bridge the gap between effectiveness and transparency in information retrieval, fostering reproducibility and further research in the field.

cs.CL↗

arXiv2025

Memory-Aware and Uncertainty-Guided Retrieval for Multi-Hop Question Answering

Multi-hop question answering (QA) requires models to retrieve and reason over multiple pieces of evidence. While Retrieval-Augmented Generation (RAG) has made progress in this area, existing methods often suffer from two key limitations: (1) fixed or overly frequent retrieval steps, and (2) ineffective use of previously retrieved knowledge. We propose MIND (Memory-Informed and INteractive Dynamic RAG), a framework that addresses these challenges through: (i) prompt-based entity extraction to identify reasoning-relevant elements, (ii) dynamic retrieval triggering based on token-level entropy and attention signals, and (iii) memory-aware filtering, which stores high-confidence facts across reasoning steps to enable consistent multi-hop generation.

cs.CL↗

arXiv2024

Mitigating the Risk of Health Inequity Exacerbated by Large Language Models

Recent advancements in large language models have demonstrated their potential in numerous medical applications, particularly in automating clinical trial matching for translational research and enhancing medical question answering for clinical decision support. However, our study shows that incorporating non decisive sociodemographic factors such as race, sex, income level, LGBT+ status, homelessness, illiteracy, disability, and unemployment into the input of LLMs can lead to incorrect and harmful outputs for these populations. These discrepancies risk exacerbating existing health disparities if LLMs are widely adopted in healthcare. To address this issue, we introduce EquityGuard, a novel framework designed to detect and mitigate the risk of health inequities in LLM based medical applications. Our evaluation demonstrates its efficacy in promoting equitable outcomes across diverse populations.

cs.CL↗

arXiv2024

RAG-RLRC-LaySum at BioLaySumm: Integrating Retrieval-Augmented Generation and Readability Control for Layman Summarization of Biomedical Texts

This paper introduces the RAG-RLRC-LaySum framework, designed to make complex biomedical research understandable to laymen through advanced Natural Language Processing (NLP) techniques. Our Retrieval Augmented Generation (RAG) solution, enhanced by a reranking method, utilizes multiple knowledge sources to ensure the precision and pertinence of lay summaries. Additionally, our Reinforcement Learning for Readability Control (RLRC) strategy improves readability, making scientific content comprehensible to non-specialists. Evaluations using the publicly accessible PLOS and eLife datasets show that our methods surpass Plain Gemini model, demonstrating a 20% increase in readability scores, a 15% improvement in ROUGE-2 relevance scores, and a 10% enhancement in factual accuracy. The RAG-RLRC-LaySum framework effectively democratizes scientific knowledge, enhancing public engagement with biomedical discoveries.

cs.CL↗

arXiv2024

Effects of Different Prompts on the Quality of GPT-4 Responses to Dementia Care Questions

Evidence suggests that different prompts lead large language models (LLMs) to generate responses with varying quality. Yet, little is known about prompts' effects on response quality in healthcare domains. In this exploratory study, we address this gap, focusing on a specific healthcare domain: dementia caregiving. We first developed an innovative prompt template with three components: (1) system prompts (SPs) featuring 4 different roles; (2) an initialization prompt; and (3) task prompts (TPs) specifying different levels of details, totaling 12 prompt combinations. Next, we selected 3 social media posts containing complicated, real-world questions about dementia caregivers' challenges in 3 areas: memory loss and confusion, aggression, and driving. We then entered these posts into GPT-4, with our 12 prompts, to generate 12 responses per post, totaling 36 responses. We compared the word count of the 36 responses to explore potential differences in response length. Two experienced dementia care clinicians on our team assessed the response quality using a rating scale with 5 quality indicators: factual, interpretation, application, synthesis, and comprehensiveness (scoring range: 0-5; higher scores indicate higher quality).

cs.CL↗