Zhikun Xu

Zhaonan Li, Kyle R. Chickering, Bangzheng Li et al.

A useful test of visual concept learning is not just whether a model can recognize a concept in a single image, but whether it can preserve and manipulate concept-level properties under transformation and transfer them to new scenes. We introduce VisAnalog, a controlled suite for this setting on natural images. Each example instantiates $A\!:\!B::C\!:\,?$: images $B$ and a hidden target image $D$ are produced by applying the same deterministic transformation sequence to source images $A$ and $C$. Given $A$, $B$, and $C$, a model must answer a multiple-choice question about $D$. The benchmark contains 617 human-validated questions spanning one- to four-step transformations such as zoom, quadrant swap, rotation, flip, and hue rotation. Across strong proprietary and open-source VLMs, end-to-end accuracy is substantially lower than oracle accuracy when $D$ is directly shown, and degrades sharply as transformation depth increases, while human performance remains near the ceiling. A program-conditioned evaluation further separates failures of relation inference from failures of transformation application, showing that inferring the visual relation from $A \rightarrow B$ is the dominant bottleneck, with additional application errors emerging on harder multi-step cases. The dataset is publicly available at https://huggingface.co/datasets/zli99/VisAnalog.

Zhikun Xu, Yu Feng, Jacob Dineen et al.

Large language model agents trained with reinforcement learning (RL) often learn brittle, task-specific shortcuts. We hypothesize that agents generalize better when their successful trajectories are structurally compressible, decomposed into a small set of reusable abstract patterns. To formalize this, we introduce ReuseRL, which grounds agentic RL in the Minimum Description Length (MDL) principle. ReuseRL extracts a shared skill dictionary from successful trajectories and augments the RL objective with a segmentation cost, explicitly penalizing idiosyncratic behaviors that encode poorly. We prove a PAC-Bayes generalization bound for this compression penalty. Across ALFWorld, TextWorld-Cooking, and Countdown-Stepwise, ReuseRL improves in- and out-of-distribution success over vanilla GRPO and strong round-length baselines.

Zhaonan Li, Shijie Lu, Fei Wang et al.

End-to-end Vision-language Models (VLMs) often answer visual questions by exploiting spurious correlations instead of causal visual evidence, and can become more shortcut-prone when fine-tuned. We introduce VISTA (Visual-Information Separation for Text-based Analysis), a modular framework that decouples perception from reasoning via an explicit information bottleneck. A frozen VLM sensor is restricted to short, objective perception queries, while a text-only LLM reasoner decomposes each question, plans queries, and aggregates visual facts in natural language. This controlled interface defines a reward-aligned environment for training unbiased visual reasoning with reinforcement learning. Instantiated with Qwen2.5-VL and Llama3.2-Vision sensors, and trained with GRPO from only 641 curated multi-step questions, VISTA significantly improves robustness to real-world spurious correlations on SpuriVerse (+16.29% with Qwen-2.5-VL-7B and +6.77% with Llama-3.2-Vision-11B), while remaining competitive on MMVP and a balanced SeedBench subset. VISTA transfers robustly across unseen VLM sensors and is able to recognize and recover from VLM perception failures. Human analysis further shows that VISTA's reasoning traces are more neutral, less reliant on spurious attributes, and more explicitly grounded in visual evidence than end-to-end VLM baselines.

Yulong Yang, Zhikun Xu, Yaojun Li et al.

When the color distribution of input images changes at inference, the performance of conventional neural network architectures drops considerably. A few researchers have begun to incorporate prior knowledge of color geometry in neural network design. These color equivariant architectures have modeled hue variation with 2D rotations, and saturation and luminance transformations as 1D translations. While this approach improves neural network robustness to color variations in a number of contexts, we find that approximating saturation and luminance (interval valued quantities) as 1D translations introduces appreciable artifacts. In this paper, we introduce a color equivariant architecture that is truly equivariant. Instead of approximating the interval with the real line, we lift values on the interval to values on the circle (a double-cover) and build equivariant representations there. Our approach resolves the approximation artifacts of previous methods, improves interpretability and generalizability, and achieves better predictive performance than conventional and equivariant baselines on tasks such as fine-grained classification and medical imaging tasks. Going beyond the context of color, we show that our proposed lifting can also extend to geometric transformations such as scale.

Jacob Dineen, Aswin RRV, Zhikun Xu et al.

Co-evolutionary self-play, where one language model generates problems and another solves them, promises autonomous curriculum learning without human supervision. In practice, the proposer quickly converges to a narrow distribution of problems that satisfy the reward function. This diversity collapse renders the curriculum uninformative for the solver, stalling the co-evolutionary loop. We introduce vocabulary dropout, a random mask applied to the proposer's output logits during both policy training and curriculum generation, as a lightweight mechanism to sustain diversity. The mask is hard and non-stationary, preventing the proposer from locking into fixed token sequences. Training Qwen3-4B and Qwen3-8B on mathematical reasoning via R-Zero, we find that vocabulary dropout sustains proposer diversity across lexical, semantic, and functional metrics throughout training, and yields solver improvements averaging +4.4 points at 8B, with the largest gains on competition-level benchmarks. Our findings suggest that explicit action-space constraints, analogous to the structural role that game rules play in classical self-play, can help sustain productive co-evolution in language. Vocabulary dropout is one simple instantiation of this principle.

Zijun Gao, Zhikun Xu, Xiao Ye et al.

Large language models (LLMs) often solve challenging math exercises yet fail to apply the concept right when the problem requires genuine understanding. Popular Reinforcement Learning with Verifiable Rewards (RLVR) pipelines reinforce final answers but provide little fine-grained conceptual signal, so models improve at pattern reuse rather than conceptual applications. We introduce CORE (Concept-Oriented REinforcement), an RL training framework that turns explicit concepts into a controllable supervision signal. Starting from a high-quality, low-contamination textbook resource that links verifiable exercises to concise concept descriptions, we run a sanity probe showing LLMs can restate definitions but fail concept-linked quizzes, quantifying the conceptual reasoning gap. CORE then (i) synthesizes concept-aligned quizzes, (ii) injects brief concept snippets during rollouts to elicit concept-primed trajectories, and (iii) reinforces conceptual reasoning via trajectory replacement after group failures, a lightweight forward-KL constraint that aligns unguided with concept-primed policies, or standard GRPO directly on concept-aligned quizzes. Across several models, CORE delivers consistent gains over vanilla and SFT baselines on both in-domain concept-exercise suites and diverse out-of-domain math benchmarks. CORE unifies direct training on concept-aligned quizzes and concept-injected rollouts under outcome regularization. It provides fine-grained conceptual supervision that bridges problem-solving competence and genuine conceptual reasoning, while remaining algorithm- and verifier-agnostic.

Cunxiang Wang, Sirui Cheng, Qipeng Guo et al.

This study focuses on the evaluation of the Open Question Answering (Open-QA) task, which can directly estimate the factuality of large language models (LLMs). Current automatic evaluation methods have shown limitations, indicating that human evaluation still remains the most reliable approach. We introduce a new task, Evaluating QA Evaluation (QA-Eval) and the corresponding dataset EVOUNA, designed to assess the accuracy of AI-generated answers in relation to standard answers within Open-QA. Our evaluation of these methods utilizes human-annotated results to measure their performance. Specifically, the work investigates methods that show high correlation with human evaluations, deeming them more reliable. We also discuss the pitfalls of current methods and methods to improve LLM-based evaluators. We believe this new QA-Eval task and corresponding dataset EVOUNA will facilitate the development of more effective automatic evaluation tools and prove valuable for future research in this area. All resources are available at \url{https://github.com/wangcunxiang/QA-Eval} and it is under the Apache-2.0 License.

Zhikun Xu, Yinghui Li, Ruixue Ding et al.

How to better evaluate the capabilities of Large Language Models (LLMs) is the focal point and hot topic in current LLMs research. Previous work has noted that due to the extremely high cost of iterative updates of LLMs, they are often unable to answer the latest dynamic questions well. To promote the improvement of Chinese LLMs' ability to answer dynamic questions, in this paper, we introduce CDQA, a Chinese Dynamic QA benchmark containing question-answer pairs related to the latest news on the Chinese Internet. We obtain high-quality data through a pipeline that combines humans and models, and carefully classify the samples according to the frequency of answer changes to facilitate a more fine-grained observation of LLMs' capabilities. We have also evaluated and analyzed mainstream and advanced Chinese LLMs on CDQA. Extensive experiments and valuable insights suggest that our proposed CDQA is challenging and worthy of more further study. We believe that the benchmark we provide will become one of the key data resources for improving LLMs' Chinese question-answering ability in the future.

Yinghui Li, Jiayi Kuang, Haojing Huang et al.

Leveraging mathematical Large Language Models (LLMs) for proof generation is a fundamental topic in LLMs research. We argue that the ability of current LLMs to prove statements largely depends on whether they have encountered the relevant proof process during training. This reliance limits their deeper understanding of mathematical theorems and related concepts. Inspired by the pedagogical method of "proof by counterexamples" commonly used in human mathematics education, our work aims to enhance LLMs' ability to conduct mathematical reasoning and proof through counterexamples. Specifically, we manually create a high-quality, university-level mathematical benchmark, CounterMATH, which requires LLMs to prove mathematical statements by providing counterexamples, thereby assessing their grasp of mathematical concepts. Additionally, we develop a data engineering framework to automatically obtain training data for further model improvement. Extensive experiments and detailed analyses demonstrate that CounterMATH is challenging, indicating that LLMs, such as OpenAI o1, have insufficient counterexample-driven proof capabilities. Moreover, our exploration into model training reveals that strengthening LLMs' counterexample-driven conceptual reasoning abilities is crucial for improving their overall mathematical capabilities. We believe that our work offers new perspectives on the community of mathematical LLMs.

Jacob Dineen, Aswin RRV, Qin Liu et al.

Alignment of large language models (LLMs) with principles like helpfulness, honesty, and harmlessness typically relies on scalar rewards that obscure which objectives drive the training signal. We introduce QA-LIGN, which decomposes monolithic rewards into interpretable principle-specific evaluations through structured natural language programs. Models learn through a draft, critique, and revise pipeline, where symbolic evaluation against the rubrics provides transparent feedback for both initial and revised responses during GRPO training. Applied to uncensored Llama-3.1-8B-Instruct, QA-LIGN reduces attack success rates by up to 68.7% while maintaining a 0.67% false refusal rate, achieving Pareto optimal safety-helpfulness performance and outperforming both DPO and GRPO with state-of-the-art reward models given equivalent training. These results demonstrate that making reward signals interpretable and modular improves alignment effectiveness, suggesting transparency enhances LLM safety.

Jiayi Kuang, Haojing Huang, Yinghui Li et al.

Large Language Models (LLMs) have demonstrated outstanding performance in mathematical reasoning capabilities. However, we argue that current large-scale reasoning models primarily rely on scaling up training datasets with diverse mathematical problems and long thinking chains, which raises questions about whether LLMs genuinely acquire mathematical concepts and reasoning principles or merely remember the training data. In contrast, humans tend to break down complex problems into multiple fundamental atomic capabilities. Inspired by this, we propose a new paradigm for evaluating mathematical atomic capabilities. Our work categorizes atomic abilities into two dimensions: (1) field-specific abilities across four major mathematical fields, algebra, geometry, analysis, and topology, and (2) logical abilities at different levels, including conceptual understanding, forward multi-step reasoning with formal math language, and counterexample-driven backward reasoning. We propose corresponding training and evaluation datasets for each atomic capability unit, and conduct extensive experiments about how different atomic capabilities influence others, to explore the strategies to elicit the required specific atomic capability. Evaluation and experimental results on advanced models show many interesting discoveries and inspirations about the different performances of models on various atomic capabilities and the interactions between atomic capabilities. Our findings highlight the importance of decoupling mathematical intelligence into atomic components, providing new insights into model cognition and guiding the development of training strategies toward a more efficient, transferable, and cognitively grounded paradigm of "atomic thinking".

Xiao Ye, Shaswat Shrivastava, Zhaonan Li et al.

Large language models excel at many tasks but still struggle with consistent, robust reasoning. We introduce Cohort-based Consistency Learning (CC-Learn), a reinforcement learning framework that improves the reliability of LLM reasoning by training on cohorts of similar questions derived from shared programmatic abstractions. To enforce cohort-level consistency, we define a composite objective combining cohort accuracy, a retrieval bonus for effective problem decomposition, and a rejection penalty for trivial or invalid lookups that reinforcement learning can directly optimize, unlike supervised fine-tuning. Optimizing this reward guides the model to adopt uniform reasoning patterns across all cohort members. Experiments on challenging reasoning benchmarks (including ARC-Challenge and StrategyQA) show that CC-Learn boosts both accuracy and reasoning stability over pretrained and SFT baselines. These results demonstrate that cohort-level RL effectively enhances reasoning consistency in LLMs.

Zhikun Xu, Ming Shen, Jacob Dineen et al.

We introduce thoughts of words (ToW), a novel training-time data-augmentation method for next-word prediction. ToW views next-word prediction as a core reasoning task and injects fine-grained thoughts explaining what the next word should be and how it is related to the previous contexts in pre-training texts. Our formulation addresses two fundamental drawbacks of existing next-word prediction learning schemes: they induce factual hallucination and are inefficient for models to learn the implicit reasoning processes in raw texts. While there are many ways to acquire such thoughts of words, we explore the first step of acquiring ToW annotations through distilling from larger models. After continual pre-training with only 70K ToW annotations, we effectively improve models' reasoning performances by 7% to 9% on average and reduce model hallucination by up to 10%. At the same time, ToW is entirely agnostic to tasks and applications, introducing no additional biases on labels or semantics.

Ming Shen, Zhikun Xu, Jacob Dineen et al.

Large language models (LLMs) are typically pretrained with next-word prediction (NWP), which yields strong surface fluency but places limited pressure on models to form explicit reasoning before emitting tokens. We study whether shifting the supervision signal can better elicit explicit reasoning and, more broadly, strengthen models' general reasoning capability. We present BOttlenecked next-Word prediction (BOW), a RL formulation of NWP that inserts an intermediate reasoning bottleneck. Instead of predicting the next word directly from context, the policy model must first generate a next-word reasoning trajectory. A frozen scorer then assigns this trajectory a soft, distributional reward equal to the probability of the gold next token conditioned solely on the trajectory to guide the RL optimization. We also propose an optional L1-style regularizer on the reward to discourage "name-the-answer" shortcuts. Across ten benchmarks, a brief BOW adaptation phase on Qwen2.5-7B-Instruct and Llama3.1-8B-Instruct improves zero-shot reasoning and outperforms strong continual-pretraining baselines, including an RL variant with a hard, binary reward and a supervised finetuning approach with augmented data, by nearly 5% on average, while achieving the top result in 7 of 10 intrinsic NWP evaluations. These results indicate that BOW is a viable alternative to vanilla NWP, inducing explicit next-word reasoning and strengthening general reasoning ability.

Zhikun Xu, Xiaodong Yu, Ben Zhou et al.

Recent large language models (LLMs) perform strongly on mathematical benchmarks yet often misapply lemmas, importing conclusions without validating assumptions. We formalize lemma$-$judging as a structured prediction task: given a statement and a candidate lemma, the model must output a precondition check and a conclusion$-$utility check, from which a usefulness decision is derived. We present RULES, which encodes this specification via a two$-$section output and trains with reinforcement learning plus section$-$aware loss masking to assign penalty to the section responsible for errors. Training and evaluation draw on diverse natural language and formal proof corpora; robustness is assessed with a held$-$out perturbation suite; and end$-$to$-$end evaluation spans competition$-$style, perturbation$-$aligned, and theorem$-$based problems across various LLMs. Results show consistent in$-$domain gains over both a vanilla model and a single$-$label RL baseline, larger improvements on applicability$-$breaking perturbations, and parity or modest gains on end$-$to$-$end tasks; ablations indicate that the two$-$section outputs and section$-$aware reinforcement are both necessary for robustness.

Xiao Ye, Jacob Dineen, Zhaonan Li et al.

Medical Large language models achieve strong scores on standard benchmarks; however, the transfer of those results to safe and reliable performance in clinical workflows remains a challenge. This survey reframes evaluation through a levels-of-autonomy lens (L0-L3), spanning informational tools, information transformation and aggregation, decision support, and supervised agents. We align existing benchmarks and metrics with the actions permitted at each level and their associated risks, making the evaluation targets explicit. This motivates a level-conditioned blueprint for selecting metrics, assembling evidence, and reporting claims, alongside directions that link evaluation to oversight. By centering autonomy, the survey moves the field beyond score-based claims toward credible, risk-aware evidence for real clinical use.

Cunxiang Wang, Zhikun Xu, Qipeng Guo et al.

The Open-Domain Question Answering (ODQA) task involves retrieving and subsequently generating answers from fine-grained relevant passages within a database. Current systems leverage Pretrained Language Models (PLMs) to model the relationship between questions and passages. However, the diversity in surface form expressions can hinder the model's ability to capture accurate correlations, especially within complex contexts. Therefore, we utilize Abstract Meaning Representation (AMR) graphs to assist the model in understanding complex semantic information. We introduce a method known as Graph-as-Token (GST) to incorporate AMRs into PLMs. Results from Natural Questions (NQ) and TriviaQA (TQ) demonstrate that our GST method can significantly improve performance, resulting in up to 2.44/3.17 Exact Match score improvements on NQ/TQ respectively. Furthermore, our method enhances robustness and outperforms alternative Graph Neural Network (GNN) methods for integrating AMRs. To the best of our knowledge, we are the first to employ semantic graphs in ODQA.