arXiv subjects

Jiachun Li

Publications and source records attributed to Jiachun Li.

At least 19 recordsLinked to original sources

arXiv2026

An Isotropy-Preserving Spectral Cap for Muon: Theory and Three Case Studies

Muon and related matrix-sign optimizers are increasingly used to pre-train large language models, but their effect on the internal geometry of individual weight matrices is not well understood. This preliminary report proposes a unified framework built on a single idealizing assumption -- exact scale invariance of the loss under weight rescaling, which holds approximately in normalization-heavy networks. Under this assumption, plain SGD carries a built-in 1/||W|| brake on its update size, whereas Muon's matrix-sign step removes that brake, so both the Frobenius and spectral norms drift outward faster (t^{1/2} versus t^{1/4}). We further observe that the spectral-norm perturbation has a non-negative second-order term. This implies that a lightweight "spectral cap" -- which projects out only the first-order growth of the single top singular direction from each update -- can control the output covariance W K_X W^T without freezing training: the weight keeps learning through non-top directions, top-direction rotation, and top switching. We relate this cap to the min-entropy (H-infinity) of the singular-value spectrum. We then study three systems trained with Muon: a nanoGPT feed-forward projection, a 64-expert mixture-of-experts router, and the query/key projections of a bf16 FlashAttention block. In each case the cap increases isotropy and, at the margins -- a router collapsing to a single expert, and the near-divergence of one attention head -- prevents a concrete failure, while leaving validation loss essentially unchanged. We emphasize that the scale-invariance assumption is strong and that these small-scale results are preliminary; comments are welcome.

cs.LG↗

arXiv2026

Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation

Adaptive experiments for average treatment effects (ATE) require randomized allocations balancing valid inference with statistical efficiency. The oracle design is a covariate-dependent Neyman rule governed by unknown arm-conditional outcome variances. We investigate whether this sequential variance-estimation and allocation process can be amortized via in-context learning. We introduce Bayesian in-context experimenters: transformer policies trained to imitate a Bayesian posterior Neyman teacher. The teacher updates nonparametric beliefs over potential outcomes using experimental history to assign posterior Neyman treatment probabilities. This design converges to the oracle rule, supporting efficient ATE inference. Transformers constructively implement this mapping through attention-based sufficient statistics and projected gradient descent, imitating Bayesian updating for Gaussian-series priors. To address unknown outcome smoothness, we combine smoothness-indexed experimenters using a mixture-of-experts transformer. The gate acts as a hierarchical posterior over smoothness classes, concentrating on near-oracle experts. By bounding the complexity of the transformer class, we prove this amortized policy can be learned via empirical risk minimization using supervised pretraining. Experiments confirm accurate teacher imitation, adaptive allocation, and improved ATE precision over baselines.

cs.LG↗

arXiv2026

Semiparametric Efficiency in Sequential Experiments: Characterization and Design via Average Propensity

Modern experiments, including evaluations of AI-enabled services and platform interventions, often depart from independent and identically distributed (i.i.d.) sampling because assignments may be adaptive, balanced across covariates, or subject to rollout constraints such as exposure, fairness, and budget limits. This paper studies the efficiency benchmark for estimating causal targets in such sequential experiments. We show that every non-anticipating design induces an average propensity score, and we establish a semiparametric lower bound: for regular locally unbiased estimators, attainable precision is bounded by the i.i.d. efficiency benchmark evaluated at this induced score. The average propensity score thereby serves as a common benchmark and design target, allowing sequential experimental design to be viewed as choosing or learning an efficient allocation rule, with operational constraints entering through the admissible set when present. We then develop implementable batched adaptive designs that approach this benchmark through two complementary mechanisms. The first uses regression adjustment based on efficient influence functions; for general smooth estimands it attains the benchmark under standard nuisance-rate conditions, while for linear functionals of outcome means it achieves a sharp second-order rate. The second uses adaptive covariate balancing to attain the same benchmark through the assignment mechanism, enabling simple moment-based estimation. Both routes require only a small number of policy updates, making them compatible with delayed feedback and easier to monitor in operational deployments. Numerical experiments and an empirical study of AI medical-assistant evaluation demonstrate the practical efficiency gains, including in multi-treatment settings. Overall, the paper provides a unified framework for characterizing and designing efficient sequential experiments.

stat.ME↗

arXiv2026

Agentic Environment Engineering for Large Language Models: A Survey of Environment Modeling, Synthesis, Evaluation, and Application

Environments serve as interactive systems for large language model (LLM) based agents across diverse scenarios and play a crucial role in driving the continual evolution of model capabilities. Despite this importance, existing work lacks a systematic categorization and deep analysis. This paper systematically studies current researches on agentic environments from the perspective of the environment engineering lifecycle, covering their modeling, synthesis, evaluation and application. Specifically, the paper first introduces representative environments from the perspectives of eight attributes and eight domains, providing detailed analyses of their development paths and highlighting their core capabilities. Second, for automated environment synthesis, two paradigms are introduced, such as symbolic synthesis and neural synthesis. This paper also shows different environment evaluation methods in each paradigm. Thirdly, the corresponding environment applications from the perspective of agent-environment co-evolution are discussed. In specific, the paper characterizes the primary pathways for agent evolution in dynamic environments from four complementary perspectives: memory-centric experience evolution, orchestration-centric workflow evolution, trajectory-centric offline evolution, and exploration-centric online evolution. And three paradigms of environment evolution are identified, namely neural-driven, difficulty-driven, and scaling-driven approaches. At last, several promising future directions are discussed, including Environment-as-a-Service, Multi-agent Environments, and Neural-Symbolic Environments.

cs.CL↗

arXiv2026

Low Rank for Rank: Uncertainty-Aware Task-Specific LLM Ranking under Sparse Pairwise Comparisons

Pairwise human-preference platforms such as Chatbot Arena have become central to large language model (LLM) evaluation, yet reliable task-specific ranking remains challenging. Global leaderboards mask task heterogeneity, while ranking each fine-grained task independently is unstable under sparse, imbalanced comparisons. We propose a low-rank framework for task-specific LLM ranking from sparse pairwise comparisons, modeling the task-by-model ability matrix $\Theta^\star \in \mathbb{R}^{d_t \times d_m}$ as low rank so that information is shared across related tasks while task-specific differences are preserved. We first develop a max-norm ($\ell_\infty$) accurate estimator for the latent scores, combining a convex initializer with alternating-minimization refinement, and prove task-wise top-$K$ recovery guarantees under sparse sampling. Our main contribution is an uncertainty quantification framework for task-specific ranking. We construct cross-fitted one-step debiased estimators for fixed score contrasts -- such as the task-specific ability gap between two models -- yielding asymptotically valid confidence intervals that attain the semiparametric efficiency bound. We then extend the inference to the high-dimensional ranking regime, where per-task ranks and top-$K$ membership are determined by many dependent score-gap hypotheses. Using Gaussian and multiplier-bootstrap calibration, we obtain simultaneous confidence sets for per-task ranks and valid top-$K$ membership tests across many tasks and models. Experiments on synthetic data and Chatbot Arena show that low-rank sharing improves sample efficiency over independent task-wise Bradley-Terry estimation and produces tighter, better-calibrated ranking certificates, with the largest gains in the sparse regime typical of real LLM benchmarks.

stat.ME↗

arXiv2026

LLM Evaluation as Tensor Completion: Low Rank Structure and Semiparametric Efficiency

Large language model (LLM) evaluation platforms increasingly rely on pairwise human judgments. These data are noisy, sparse, and non-uniform, yet leaderboards are reported with limited uncertainty quantification. We study this as semiparametric inference for a low-rank latent score tensor observed through pairwise comparisons under Bradley-Terry-Luce-type models. This places LLM evaluation in a new tensor completion setting with structured observations, non-uniform sampling, and pairwise contrasts. Our target is a smooth functional $\psi(T^\star)$, including linear estimands such as ability gaps and nonlinear ones such as win probabilities. We derive the information operator on the low-rank tangent space, the efficient influence function, and the semiparametric efficiency bound, then construct a one-step debiased estimator with asymptotic normality. A central challenge is that the information operator is anisotropic and does not commute with the tangent-space projection, creating a bottleneck absent from isotropic models. We introduce a score-whitening method that equalizes local Fisher information and restores stable inference at the optimal sample-complexity scale. Our results provide a principled framework for uncertainty quantification in LLM evaluation and more broadly for inference on low-rank structures from pairwise data.

stat.ME↗

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

Your Microphone Array Retains Your Identity: A Robust Voice Liveness Detection System for Smart Speakers

Though playing an essential role in smart home systems, smart speakers are vulnerable to voice spoofing attacks. Passive liveness detection, which utilizes only the collected audio rather than the deployed sensors to distinguish between live-human and replayed voices, has drawn increasing attention. However, it faces the challenge of performance degradation under the different environmental factors as well as the strict requirement of the fixed user gestures. In this study, we propose a novel liveness feature, array fingerprint, which utilizes the microphone array inherently adopted by the smart speaker to determine the identity of collected audios. Our theoretical analysis demonstrates that by leveraging the circular layout of microphones, compared with existing schemes, array fingerprint achieves a more robust performance under the environmental change and user's movement. Then, to leverage such a fingerprint, we propose ARRAYID, a lightweight passive detection scheme, and elaborate a series of features working together with array fingerprint. Our evaluation on the dataset containing 32,780 audio samples and 14 spoofing devices shows that ARRAYID achieves an accuracy of 99.84%, which is superior to existing passive liveness detection schemes.

cs.CR↗

arXiv2025

Omni-Reward: Towards Generalist Omni-Modal Reward Modeling with Free-Form Preferences

Reward models (RMs) play a critical role in aligning AI behaviors with human preferences, yet they face two fundamental challenges: (1) Modality Imbalance, where most RMs are mainly focused on text and image modalities, offering limited support for video, audio, and other modalities; and (2) Preference Rigidity, where training on fixed binary preference pairs fails to capture the complexity and diversity of personalized preferences. To address the above challenges, we propose Omni-Reward, a step toward generalist omni-modal reward modeling with support for free-form preferences, consisting of: (1) Evaluation: We introduce Omni-RewardBench, the first omni-modal RM benchmark with free-form preferences, covering nine tasks across five modalities including text, image, video, audio, and 3D; (2) Data: We construct Omni-RewardData, a multimodal preference dataset comprising 248K general preference pairs and 69K instruction-tuning pairs for training generalist omni-modal RMs; (3) Model: We propose Omni-RewardModel, which includes both discriminative and generative RMs, and achieves strong performance on Omni-RewardBench as well as other widely used reward modeling benchmarks.

cs.CL↗

arXiv2025

Beyond ATE: Multi-Criteria Design for A/B Testing

In the era of large-scale AI deployment and high-stakes clinical trials, adaptive experimentation faces a ``trilemma'' of conflicting objectives: minimizing cumulative regret (welfare loss during the experiment), maximizing the estimation accuracy of heterogeneous treatment effects (CATE), and ensuring differential privacy (DP) for participants. Existing literature typically optimizes these metrics in isolation or under restrictive parametric assumptions. In this work, we study the multi-objective design of adaptive experiments in a general non-parametric setting. First, we rigorously characterize the instance-dependent Pareto frontier between cumulative regret and estimation error, revealing the fundamental statistical limits of dual-objective optimization. We propose ConSE, a sequential segmentation and elimination algorithm that adaptively discretizes the covariate space to achieve the Pareto-optimal frontier. Second, we introduce DP-ConSE, a privacy-preserving extension that satisfies Joint Differential Privacy. We demonstrate that privacy comes ``for free'' in our framework, incurring only asymptotically negligible costs to regret and estimation accuracy. Finally, we establish a robust link between experimental design and long-term utility: we prove that any policy derived from our Pareto-optimal algorithms minimizes post-experiment simple regret, regardless of the specific exploration-exploitation trade-off chosen during the trial. Our results provide a theoretical foundation for designing ethical, private, and efficient adaptive experiments in sensitive domains.

stat.ME↗

arXiv2025

MMR-V: What's Left Unsaid? A Benchmark for Multimodal Deep Reasoning in Videos

The sequential structure of videos poses a challenge to the ability of multimodal large language models (MLLMs) to locate multi-frame evidence and conduct multimodal reasoning. However, existing video benchmarks mainly focus on understanding tasks, which only require models to match frames mentioned in the question (hereafter referred to as "question frame") and perceive a few adjacent frames. To address this gap, we propose MMR-V: A Benchmark for Multimodal Deep Reasoning in Videos. The benchmark is characterized by the following features. (1) Long-range, multi-frame reasoning: Models are required to infer and analyze evidence frames that may be far from the question frame. (2) Beyond perception: Questions cannot be answered through direct perception alone but require reasoning over hidden information. (3) Reliability: All tasks are manually annotated, referencing extensive real-world user understanding to align with common perceptions. (4) Confusability: Carefully designed distractor annotation strategies to reduce model shortcuts. MMR-V consists of 317 videos and 1,257 tasks. Our experiments reveal that current models still struggle with multi-modal reasoning; even the best-performing model, Gemini-2.5-pro, achieves only 64.3% accuracy. Additionally, current reasoning enhancement strategies (Chain-of-Thought and scaling test-time compute) bring limited gains. Error analysis indicates that the CoT demanded for multi-modal reasoning differs from it in textual reasoning, which partly explains the limited performance gains. We hope that MMR-V can inspire further research into enhancing multi-modal reasoning capabilities.

cs.CV↗

arXiv2025

Fixing the Broken Compass: Diagnosing and Improving Inference-Time Reward Modeling

Inference-time scaling techniques have shown promise in enhancing the reasoning capabilities of large language models (LLMs). While recent research has primarily focused on training-time optimization, our work highlights inference-time reward model (RM)-based reasoning as a critical yet overlooked avenue. In this paper, we conduct a systematic analysis of RM behavior across downstream reasoning tasks, revealing three key limitations: (1) RM can impair performance on simple questions, (2) its discriminative ability declines with increased sampling, and (3) high search diversity undermines RM performance. To address these issues, we propose CRISP (Clustered Reward Integration with Stepwise Prefixing), a novel inference-time algorithm that clusters generated reasoning paths by final answers, aggregates reward signals at the cluster level, and adaptively updates prefix prompts to guide generation. Experimental results demonstrate that CRISP significantly enhances LLM reasoning performance, achieving up to 5% accuracy improvement over other RM-based inference methods and an average of 10% gain over advanced reasoning models.

cs.CL↗

arXiv2025

MIRAGE: Evaluating and Explaining Inductive Reasoning Process in Language Models

Inductive reasoning is an essential capability for large language models (LLMs) to achieve higher intelligence, which requires the model to generalize rules from observed facts and then apply them to unseen examples. We present MIRAGE, a synthetic dataset that addresses the limitations of previous work, specifically the lack of comprehensive evaluation and flexible test data. In it, we evaluate LLMs' capabilities in both the inductive and deductive stages, allowing for flexible variation in input distribution, task scenario, and task difficulty to analyze the factors influencing LLMs' inductive reasoning. Based on these multi-faceted evaluations, we demonstrate that the LLM is a poor rule-based reasoner. In many cases, when conducting inductive reasoning, they do not rely on a correct rule to answer the unseen case. From the perspectives of different prompting methods, observation numbers, and task forms, models tend to consistently conduct correct deduction without correct inductive rules. Besides, we find that LLMs are good neighbor-based reasoners. In the inductive reasoning process, the model tends to focus on observed facts that are close to the current test example in feature space. By leveraging these similar examples, the model maintains strong inductive capabilities within a localized region, significantly improving its deductive performance.

cs.CL↗

arXiv2025

Beyond Covariance Matrix: The Statistical Complexity of Private Linear Regression

We study the statistical complexity of private linear regression under an unknown, potentially ill-conditioned covariate distribution. Somewhat surprisingly, under privacy constraints the intrinsic complexity is \emph{not} captured by the usual covariance matrix but rather its $L_1$ analogues. Building on this insight, we establish minimax convergence rates for both the central and local privacy models and introduce an Information-Weighted Regression method that attains the optimal rates. As application, in private linear contextual bandits, we propose an efficient algorithm that achieves rate-optimal regret bounds of order $\sqrt{T}+\frac{1}{\alpha}$ and $\sqrt{T}/\alpha$ under joint and local $\alpha$-privacy models, respectively. Notably, our results demonstrate that joint privacy comes at almost no additional cost, addressing the open problems posed by Azize and Basu (2024).

cs.LG↗

arXiv2024

Optimal Adaptive Experimental Design for Estimating Treatment Effect

Given n experiment subjects with potentially heterogeneous covariates and two possible treatments, namely active treatment and control, this paper addresses the fundamental question of determining the optimal accuracy in estimating the treatment effect. Furthermore, we propose an experimental design that approaches this optimal accuracy, giving a (non-asymptotic) answer to this fundamental yet still open question. The methodological contribution is listed as following. First, we establish an idealized optimal estimator with minimal variance as benchmark, and then demonstrate that adaptive experiment is necessary to achieve near-optimal estimation accuracy. Secondly, by incorporating the concept of doubly robust method into sequential experimental design, we frame the optimal estimation problem as an online bandit learning problem, bridging the two fields of statistical estimation and bandit learning. Using tools and ideas from both bandit algorithm design and adaptive statistical estimation, we propose a general low switching adaptive experiment framework, which could be a generic research paradigm for a wide range of adaptive experimental design. Through novel lower bound techniques for non-i.i.d. data, we demonstrate the optimality of our proposed experiment. Numerical result indicates that the estimation accuracy approaches optimal with as few as two or three policy updates.

stat.ML↗

arXiv2024

Fight Fire with Fire: Combating Adversarial Patch Attacks using Pattern-randomized Defensive Patches

Object detection has found extensive applications in various tasks, but it is also susceptible to adversarial patch attacks. The ideal defense should be effective, efficient, easy to deploy, and capable of withstanding adaptive attacks. In this paper, we adopt a counterattack strategy to propose a novel and general methodology for defending adversarial attacks. Two types of defensive patches, canary and woodpecker, are specially-crafted and injected into the model input to proactively probe or counteract potential adversarial patches. In this manner, adversarial patch attacks can be effectively detected by simply analyzing the model output, without the need to alter the target model. Moreover, we employ randomized canary and woodpecker injection patterns to defend against defense-aware attacks. The effectiveness and practicality of the proposed method are demonstrated through comprehensive experiments. The results illustrate that canary and woodpecker achieve high performance, even when confronted with unknown attack methods, while incurring limited time overhead. Furthermore, our method also exhibits sufficient robustness against defense-aware attacks, as evidenced by adaptive attack experiments.

cs.CV↗

arXiv2024

LINKED: Eliciting, Filtering and Integrating Knowledge in Large Language Model for Commonsense Reasoning

Large language models (LLMs) sometimes demonstrate poor performance on knowledge-intensive tasks, commonsense reasoning is one of them. Researchers typically address these issues by retrieving related knowledge from knowledge graphs or employing self-enhancement methods to elicit knowledge in LLMs. However, noisy knowledge and invalid reasoning issues hamper their ability to answer questions accurately. To this end, we propose a novel method named eliciting, filtering and integrating knowledge in large language model (LINKED). In it, we design a reward model to filter out the noisy knowledge and take the marginal consistent reasoning module to reduce invalid reasoning. With our comprehensive experiments on two complex commonsense reasoning benchmarks, our method outperforms SOTA baselines (up to 9.0% improvement of accuracy). Besides, to measure the positive and negative impact of the injected knowledge, we propose a new metric called effectiveness-preservation score for the knowledge enhancement works. Finally, through extensive experiments, we conduct an in-depth analysis and find many meaningful conclusions about LLMs in commonsense reasoning tasks.

cs.CL↗

arXiv2024

Towards Scalable GPU-Accelerated SNN Training via Temporal Fusion

Drawing on the intricate structures of the brain, Spiking Neural Networks (SNNs) emerge as a transformative development in artificial intelligence, closely emulating the complex dynamics of biological neural networks. While SNNs show promising efficiency on specialized sparse-computational hardware, their practical training often relies on conventional GPUs. This reliance frequently leads to extended computation times when contrasted with traditional Artificial Neural Networks (ANNs), presenting significant hurdles for advancing SNN research. To navigate this challenge, we present a novel temporal fusion method, specifically designed to expedite the propagation dynamics of SNNs on GPU platforms, which serves as an enhancement to the current significant approaches for handling deep learning tasks with SNNs. This method underwent thorough validation through extensive experiments in both authentic training scenarios and idealized conditions, confirming its efficacy and adaptability for single and multi-GPU systems. Benchmarked against various existing SNN libraries/implementations, our method achieved accelerations ranging from $5\times$ to $40\times$ on NVIDIA A100 GPUs. Publicly available experimental codes can be found at https://github.com/EMI-Group/snn-temporal-fusion.

cs.AI↗