Academic Intelligence · Curated Daily

探索全球前沿学术脉络

AcademicHub 汇聚顶级期刊与预印本平台的实时文献。定制您的专属科研雷达,利用大语言模型自动生成交叉领域文献分析简报。

01.
arXiv (quant-ph) 2026-06-16

A Gauge-Covariant Geometric Framework for Non-Hermitian Quantum Systems

arXiv:2606.15922v1 Announce Type: new Abstract: We develop a comprehensive, gauge-covariant geometric framework for non-Hermitian quantum systems in the quasi-Hermitian regime, that is, the region of parameter space where the non-Hermitian Hamiltonian admits a real spectrum and a positive-definite metric operator. We build this framework by elevating the Dyson map to a central geometric object. This map is the transformation that converts a non-Hermitian Hamiltonian into an equivalent Hermitian one. From it we construct the Dyson connection and decompose it into Hermitian and anti-Hermitian parts, identified respectively as {\it stretching } and {\it rotation } components. This decomposition cleanly separates the genuine physical metric deformations from the unitary gauge redundancies. Working with manifestly gauge-covariant states, we then derive the complex non-Hermitian Berry phase and the quantum geometric tensor (QGT), and show that the non-Hermitian geometric curvature originates from the non-commutativity of the stretching components at the operator level. We further analyse the geometric singularities near an exceptional point (EP) and uncover a distinct hierarchy of divergences. For a general two-level non-Hermitian model, the quantum metric tensor (QMT) exhibits a leading-order divergence $\sim |\epsilon_\mu|^{-2}$, while the Berry curvature shows a weaker, subleading divergence $\sim |\epsilon_\mu|^{-3/2}$, with $\epsilon_\mu$ denoting the parameter displacement from the EP along an individual parameter axis $\mu$. Finally, we examine physical realizations of this model, including the non-Hermitian Su–Schrieffer–Heeger (SSH) and Hatano–Nelson (HN) models, where exact analytical results confirm the predicted critical scaling laws and illustrate the metric-deformation-driven non-Hermitian geometries.

02.
arXiv (math.PR) 2026-06-17

Large deviation principle for friendship-biases in Galton–Watson trees

arXiv:2606.17381v1 Announce Type: new Abstract: In this paper we consider the friendship-bias of the vertices in an infinite rooted Galton–Watson tree. The friendship-bias of a vertex is the difference between the average degree of the neighbours of the vertex and the degree of the vertex itself. A vertex is said to be of type $\chi \in S$, with $S = \{-,0,+\}$, when its friendship-bias is, respectively, strictly negative, zero or strictly positive. We consider the fractions $f_l^\chi$ of vertices of type $\chi \in S$ along a random downward path up to branching depth $l \in \mathbb{N}$ and derive a large deviation principle (LDP) for the triple $(f_l^\chi)_{\chi \in S}$ as $l\to\infty$. The branching depth of a vertex counts the number of branchings that occur along the path that connects the vertex to the root of the tree. The rate in the LDP is $l$, while the rate function in the LDP is identified in terms of a variational formula minimising a relative entropy under a linear constraint. We focus on the case of binary branching, for which the rate function is already quite involved. We identify the qualitative properties of the rate function and show how it can be computed numerically. We briefly indicate how to proceed for more general branching and for vertex types along a tree consisting of a finite number of random downward paths. Our paper is the first to consider large deviations of vertex types.

03.
arXiv (CS.CV) 2026-06-12

MPMWorlds: Material-Point-Method Simulations for Inferring and Extrapolating Physical Dynamics

To study the ability to infer physical dynamics from videos and extrapolate them forward in time, we assemble a dataset of 2D Material Point Method (MPM) physical simulations covering rich physical phenomena such as deformable objects, fluids, kinetic objects, and emitters. We study code generation and video diffusion approaches on this dataset, identifying their strengths and weaknesses by varying the amount of physically relevant side information. The code generation model, beyond giving a working demonstration of automatic synthesis of MPM simulations, reveals that such an approach struggles with inferring physical parameters from visual input, but relative to video diffusion, produces physically and temporally stable extrapolations forward in time, while the video diffusion model more strongly identifies geometric properties from visual input but produces physically implausible extrapolations.

04.
arXiv (CS.CV) 2026-06-16

Cross-modal Identity Mapping: Minimizing Information Loss in Modality Conversion via Reinforcement Learning

Large Vision-Language Models (LVLMs) often omit or misrepresent critical visual content in generated image captions. Minimizing such information loss will force LVLMs to focus on image details to generate precise descriptions. However, measuring information loss during modality conversion is inherently challenging due to the modal gap between visual content and text output. In this paper, we argue that the quality of an image caption is positively correlated with the similarity between images retrieved via text search using that caption. Based on this insight, we further propose Cross-modal Identity Mapping (CIM), a reinforcement learning framework that enhances image captioning without requiring additional annotations. Specifically, the method quantitatively evaluates the information loss from two perspectives: Gallery Representation Consistency and Query-gallery Image Relevance. Supervised under these metrics, LVLM minimizes information loss and aims to achieve identity mapping from images to captions. The experimental results demonstrate the superior performance of our method in image captioning, even when compared with Supervised Fine-Tuning. Particularly, on the COCO-LN500 benchmark, CIM achieves a 20% improvement in relation reasoning on Qwen2.5-VL-7B.

05.
arXiv (CS.LG) 2026-06-11

IAPO: Input Attribution-Aware Policy Optimization for Tool Use in Small Multimodal Agents

arXiv:2606.11652v1 Announce Type: new Abstract: This paper investigates reinforcement learning (RL) methods for improving tool-calling capabilities in multimodal small language model (SLM) agents. While existing works have explored various reward designs to improve agentic tool-calling ability, these approaches face inherent limitations for SLM training, especially under multimodal scenarios. First, many existing methods evaluate tool use correctness through exact matching against certain ground-truth or predefined formats. However, this assumption is often unsuitable for multimodal tasks, where multiple tool use paths may be valid and annotated tool trajectories are typically unavailable. Second, such sparse and brittle binary rewards provide little guidance on how to improve the underlying decision process, making them particularly difficult for multimodal SLM to learn from. To address these issues, we propose Input Attribution-Aware Policy Optimization (IAPO), an RL algorithm for improving tool use in multimodal SLM by aligning the model's attribution across input components with that of a stronger teacher. Experiments on Qwen2.5-VL-3B show that the proposed method improves visual question answering accuracy by an average of 3% across six test sets compared with existing visual tool use work, by helping the model attend to the most relevant input evidence.

06.
arXiv (CS.LG) 2026-06-11

Bypassing Prompt Guards in Production with Controlled-Release Prompting

arXiv:2510.01529v4 Announce Type: replace Abstract: Ball et al. recently established that prompt filtering for AI alignment faces a fundamental barrier: under standard cryptographic assumptions, no filter running significantly faster than the protected model can universally distinguish adversarial prompts from benign ones. We investigate whether this impossibility result translates to real-world vulnerabilities in deployed large language model (LLM) systems. We answer affirmatively by introducing controlled-release prompting, a practical instantiation of the theoretical framework that exploits the resource asymmetry between lightweight input filters and the main models they protect. Unlike the theoretical construction, our attack does not require model modification: it generates malicious prompts that are indecipherable by any bounded filter yet remain tractable to the target LLM. We find our attack to be successful on four major chat platforms (Google Gemini, DeepSeek Chat, xAI Grok, and Mistral Le Chat) where baseline methods fail. Additionally, we apply our attack to extract copyrighted data from Gemini. Finally, we provide a systematic evaluation of 14 open-weight prompt guard models, revealing that even reasoning-capable filters cannot reliably detect our attack without incurring prohibitive resource overhead.

07.
arXiv (CS.CV) 2026-06-19

Evaluation of Image Matching for Art Skills Assessment

While some individuals possess a natural talent for drawing, mastering this skill requires dedicated training and practice. Determining one's skill in the art of drawing requires proper comprehensive assessment. In this paper, we propose a method to measure drawing skill by by matching the hand-drawn image with the original template. Existing techniques often involve complex processes. However, advancements in computer vision allow us to train computers to perform these comparisons at a human-like level, thereby resolving the tedious and overwhelming traditional process. Using computer vision applications, determining image similarity involves identifying the level of similarities in an image with a reference image. We have implemented and analyzed the SIFT feature and Siamese network to measure image similarity. Our results indicate that it is feasible to assess art skill levels. Through feature analysis, we found that SIFT-based key point matching provides a more effective means of detecting drawing skills.

08.
arXiv (CS.LG) 2026-06-16

HRIR-Former: Grid-Free Time-Domain Reconstruction of Head-Related Impulse Responses with a Spatially Encoded Transformer

arXiv:2603.27998v2 Announce Type: replace-cross Abstract: Individualized head-related impulse responses (HRIRs) enable binaural rendering, but dense per-listener measurements are costly. We address HRIR spatial up-sampling from sparse per-listener measurements: given a few measured HRIRs for a listener, predict HRIRs at unmeasured target directions. Prior learning methods often work in the frequency domain, rely on minimum-phase assumptions or separate timing models, and use a fixed direction grid, which can degrade temporal fidelity and spatial continuity. We propose HRIR-Former, a time-domain, grid-free binaural Transformer for reconstructing HRIRs at arbitrary directions from sparse inputs. It uses sinusoidal spatial features, a Conv1D refinement module, and auxiliary interaural time difference (ITD) and interaural level difference (ILD) heads. On SONICOM, it improves normalized mean squared error (NMSE), cosine distance, and ITD/ILD errors over prior methods; ablations validate modules and show minimum-phase preprocessing is unnecessary.

09.
Nature (Science) 2026-06-10

Daily briefing: Ancient ground squirrels ate like ‘zombies of the Pleistocene’

作者:

Evidence from fossilized poo reveals the diverse diet of ancient ground squirrels. Plus, the science behind the peptide craze and our innate tendency to wander anticlockwise. Evidence from fossilized poo reveals the diverse diet of ancient ground squirrels. Plus, the science behind the peptide craze and our innate tendency to wander anticlockwise.

10.
arXiv (CS.CL) 2026-06-12

Can Factual Opinions Be Edited (Manipulated) in Large Language Models?

Large Language Models (LLMs) are increasingly integrated into various domains, making knowledge editing techniques crucial yet potentially hazardous. Current editing methods primarily target atomic facts, overlooking the significant risks associated with manipulating factual opinions, e.g., documented stances of public figures on societal issues. Such manipulation could reshape public images, influence elections, and alter societal views. To systematically assess this threat, we introduce the Factual Opinion Editing with Evidence (FOE) benchmark, which encompasses 261 public figures, 19 issue categories, and 2,178 complete opinion records. Our evaluations demonstrate that current editing techniques struggle significantly with factual opinions, often achieving only superficial changes while failing to preserve consistency between the edited opinion and the supporting evidence generated by the model. To address this limitation, we further propose a simple yet effective Self-Generated Evidence-Aligned method that achieves opinion-evidence alignment without relying on explicit instructions. Together, our benchmark and method provide a foundation for understanding the emerging security implications of factual opinion editing in LLMs.

11.
Nature (Science) 2026-06-12

‘Student Geng’ ignites research-integrity scandal in China after calling out senior academics<b> </b>

作者:

Video blogger’s viral accusations of data manipulation in Nature journals have sparked intense debate and speedy institutional investigations. Video blogger’s viral accusations of data manipulation in Nature journals have sparked intense debate and speedy institutional investigations.

12.
arXiv (CS.CL) 2026-06-19

IHUBERT: Vector-Based Semantic Deduplication and Domain-Balanced Pretraining for Persian Resources

Persian pretrained language models (PLMs) are still limited by the scarcity of large-scale, high-quality pretraining corpora and by insufficient evaluation beyond standard classification and NER tasks. We present IHUBERT, a monolingual Persian PLM trained from scratch with the RoBERTa-base encoder (125M parameters) on a 45 GB curated subset of the Sepahr-Danesh collection (about 7-8B tokens). To improve corpus quality and reduce redundancy, we employ a multi-stage preprocessing pipeline that includes normalization, exact and near-duplicate removal, anonymization, and vector-database-based semantic deduplication for distribution balancing control across domains and registers. We additionally train a 139k-vocabulary BPE tokenizer on the full pretraining corpus to better capture Persian morphology and orthographic variation. IHUBERT is evaluated on seven Persian NLU benchmarks covering NER, sentiment analysis, topic classification, NLI, extractive question answering, and relation extraction, using task-standard metrics (entity-level F1, Macro-F1, EM/F1). IHUBERT achieves its strongest gains on extractive QA, ranking first on both PQuAD (F1 88.3542) and ParsiNLU-RC (F1 49.0987), and attains the best result on FarsTail (Macro-F1 0.8350). On NER and topic classification, it remains competitive (e.g., 0.8308 F1 on ParsTwiNER; 0.7953 Macro-F1 on DigiMag), while relation extraction remains the main remaining gap (0.6684 Macro-F1 on PERLEX). A controlled tokenizer ablation on the IHUBERT pretraining corpus shows that BPE yields slightly lower subword fragmentation than WordPiece at matched vocabulary size, supporting our tokenization design. Overall, IHUBERT advances Persian language modeling through semantically curated large-scale pretraining and broad evaluation across both classification and comprehension-oriented tasks.

13.
arXiv (CS.LG) 2026-06-16

Discrimination-free Insurance Pricing with Privatized Sensitive Attributes

arXiv:2504.11775v3 Announce Type: replace-cross Abstract: Fairness has become an important concern in insurance pricing as insurers increasingly rely on machine learning models to predict expected losses. At the same time, regulatory and privacy constraints often restrict insurers' ability to access or use sensitive attributes such as gender or race. Recent actuarial research addresses fairness in this context through the concept of the discrimination-free premium, which removes both the direct and indirect effects of sensitive attributes while preserving actuarial consistency. However, implementing this approach typically requires access to the sensitive attributes themselves, which may not be available in practice. This paper studies the estimation of discrimination-free insurance premiums when sensitive attributes are observed only in privatized or noise-perturbed form. We consider a multi-party data setting in which insurers observe non-sensitive attributes and outcomes, while a trusted third party holds privatized sensitive attributes generated through a privacy mechanism. Within this framework, we develop statistical methods for estimating discrimination-free premiums using only the privatized attributes. We study two settings of practical relevance: when the privacy mechanism is known and when its noise level is unknown. For both cases, we establish theoretical guarantees for the proposed estimators. Numerical experiments and empirical applications demonstrate that the proposed approach enables fair insurance pricing while respecting privacy and regulatory constraints.

14.
arXiv (CS.LG) 2026-06-16

Using Reinforcement Learning to Optimize the Global and Local Crossing Number

arXiv:2509.06108v2 Announce Type: replace-cross Abstract: Graph drawing concerns the algorithmic visualization of graphs. A good drawing of a graph is easy to read and facilitates solving tasks on the graph. Several properties have been identified to occur in good drawings of graphs. Such properties include a low number of crossings, large angles between edges, short edges, and depicting symmetries. Many of these properties are explicitly measurable metrics. This brings us to the insight that graph drawing can be seen as a game. In this paper, we study a single-player optimization game in which the player iteratively moves vertices of a straight-line graph drawing to reduce edge crossings. This game arose naturally from the automatic track of the Graph Drawing Challenge, where solutions are obtained by repeatedly performing local vertex movements. We formalize this process as a game with full information and investigate whether reinforcement learning can discover effective strategies for playing it. Our reinforcement-learning agent observes the local geometric and structural context of a vertex and selects a movement direction with the goal of reducing either the global or the local crossing number, that is, the total number of crossings or the maximum number of crossings per edge. We compare the resulting strategies to existing methods and established crossing-minimization heuristics on standard benchmark graphs. While our approach does not out-compete state-of-the-art methods for minimizing the global crossing number, it is competitive and often superior for minimizing the local crossing number.

15.
arXiv (CS.AI) 2026-06-19

Bring My Cup! Personalizing Vision-Language-Action Models with Visual Attentive Prompting

arXiv:2512.20014v3 Announce Type: replace-cross Abstract: While Vision-Language-Action (VLA) models generalize well to generic instructions, they struggle with personalized commands such as "bring my cup," where the robot must act on one specific instance among visually similar objects. We study this setting of manipulating personal objects, in which a VLA must identify and control a user-specific object unseen during training using only a few reference images. To address this challenge, we propose Visual Attentive Prompting (VAP), a simple-yet-effective training-free perceptual adapter that equips frozen VLAs with top-down selective attention. VAP treats the reference images as a non-parametric visual memory, grounds the personal object in the scene through open-vocabulary detection and embedding-based matching, and then injects this grounding as a visual prompt by highlighting the object and rewriting the instruction. We construct two simulation benchmarks, Personalized-SIMPLER and Personalized-VLABench, and a real-world tabletop benchmark to evaluate personalized manipulation across multiple robots and tasks. Experiments show that VAP consistently outperforms generic policies and token-learning baselines in both success rate and correct-object manipulation, helping to bridge the gap between semantic understanding and instance-level control.

16.
arXiv (CS.LG) 2026-06-16

Scalar-pathway fidelity improves physical accuracy in short-range equivariant interatomic potentials

arXiv:2606.15892v1 Announce Type: new Abstract: Accurate interatomic potentials enable molecular dynamics of materials, molecules, and interfaces beyond density-functional-theory length and time scales. Equivariant neural network potentials have improved the representation of local geometry. However, their deployable energy surfaces ultimately manifest through invariant scalar channels, whose aggregation and spectral resolution remain comparatively underexamined. Here we use Physics-Aware Neighborhood (PAN) pooling and Physics-Guided Spectral (PGS) mixers as controlled scalar-pathway probes: lightweight, symmetry-preserving modifications that act only on \(\ell=0\) channels while leaving the equivariant tensor backbone unchanged. Using MACE as a high-body-order mechanistic scaffold, PAN adds coordination-sensitive amplitude modulation, whereas PGS augments edge and readout scalar features with radial and tapered spectral bases. Across metallic Ag, covalent Si, a short-range ionic LiF/Li–F subset, and MD17/rMD17 molecules, this scalar-pathway correction reduces MACE force errors by 22–27\% and energy errors by 19–22\%; on systems with stress labels, stress errors decrease by 27–28\%, at approximately 5\% additional inference-FLOPs cost. Directionally consistent gains in Allegro and NequIP further indicate that the correction is portable across distinct short-range equivariant backbones, although effect sizes remain architecture-dependent. These results identify scalar-pathway fidelity as a practical design dimension for short-range equivariant interatomic potentials.

17.
arXiv (CS.CL) 2026-06-18

ToolGrad: Efficient Tool-use Dataset Generation with Textual "Gradients"

Prior work synthesizes tool-use LLM datasets by first generating a user query, followed by complex tool-use annotations like depth-first search (DFS). This leads to inevitable annotation failures and low efficiency in data generation. We introduce ToolGrad, an agentic framework that inverts this paradigm. ToolGrad first constructs valid tool-use chains through an iterative process guided by textual "gradients", and then synthesizes corresponding user queries. This "answer-first" approach led to ToolGrad-500, a dataset generated with more complex tool use, lower cost, and almost 100% pass rate. Experiments show that ToolGrad models outperform those trained on expensive baseline datasets and proprietary LLMs. The ToolGrad source code, dataset, and models are available at https://github.com/zhongyi-zhou/toolgrad.

18.
arXiv (CS.CL) 2026-06-11

Hey Chat, Can You Teach Me? Structuring Socratic Dialogue for Human Learning in the Wild

Large language models are now widely used for everyday learning, but the underlying interactions are typically unstructured chats rather than following a curriculum. Unlike formal online learning systems, these interactions carry no prior record of the student, so any estimate of what the student already knows must be inferred from the dialogue itself. We show that this gap is not closed by scaling models alone. Frontier and education-tuned LLMs perform poorly when asked to tutor a student over an extended session, because doing so requires three things at once. The tutor must sequence a curriculum, conduct Socratic dialogue, and infer the student's knowledge state from that dialogue. We propose separating these responsibilities. Given a student query, our system constructs a prerequisite knowledge graph in which subtopics are nodes and dependencies are edges, and frames tutoring as deciding which node to teach next and how many dialogue turns to spend on it before moving on. A lightweight PPO policy handles this sequencing decision, while an LLM conducts the Socratic exchange at the chosen node and returns a signal of student progress. Across held-out STEM and non-STEM topics, our PPO-paired tutor outperforms heuristic baselines, frontier general-purpose models, and a model specialised for Socratic dialogue: on both the rate at which students reach full curriculum mastery and the number of turns required. Explicit curriculum structure delivers gains that scaling the underlying model does not.

19.
bioRxiv (Bioinfo) 2026-06-19

Evaluation of analysis modes for RNA coexpression in single-cell and bulk tissue

Coexpression of transcripts presents the most common means of computational inference of transcription factor regulation, and is often combined with other data types to infer regulatory networks. With the growing popularity of single-cell approaches, there are questions about how best to extract coexpression information from the data. Recently we reported a simulation study that explored the differences among coexpression performed at different levels: across single cells (xCell, per cell type), across subjects from pseudobulked single-cell data (xSubject, per cell type), or across subjects using bulk tissue samples (xBulk). Here we test predictions made by those models using real data. We consider both preservation (consistency of coexpression findings across different levels of analysis of the same data) and replicability across independent studies, as well as biological interpretability. We find that preservation across levels is limited, indicating the choice of analysis level will affect outcomes. We show that xCell coexpression is more replicable across studies compared to xSubject. xBulk coexpression is dominated by patterns driven by variability in cellular composition and fails to capture much coexpression that is reliably detected at finer resolutions. While all modes of analysis exhibit some enrichment for known regulatory relationships, it was highest with the xCell mode. Finally, we present a case study of the effect of analysis modes on a schizophrenia-associated pattern, reinforcing the importance of analytic choices in the interpretation and replicability of coexpression analyses. Together with our modeling study, this work emphasizes the importance of understanding sources of expression covariation as they relate to the goals of the analysis, and recommend single-cell-based data with biological replicates should be the focus of attempts to infer dynamic regulatory interactions that are more likely to be replicable by others.

20.
arXiv (CS.CL) 2026-06-18

PatchWorld: Gradient-Free Optimization of Executable World Models

Text-agent environments are typically modeled as partially observable Markov decision processes (POMDPs), assuming that the simulator's latent state and transition dynamics are hidden from the agent. Yet little work has examined whether executable code can be induced to serve as a world model for prediction and planning under partial observability. We introduce PatchWorld, a gradient-free framework that turns offline trajectories into executable Python world models through counterexample-guided code repair. Instead of predicting the next observation with a black-box model, PatchWorld induces symbolic belief-state programs whose action updates can be inspected, replayed, and locally patched. Across seven AgentGym environments, PatchWorld-Simple achieves the highest code-based planning score among evaluated methods, reaching 76.4\% macro success in live one-step lookahead while invoking no LLM calls inside the world-model prediction module itself. We further find that a human-specified residual-memory bias improves surface observation fidelity but weakens decision utility. This exposes a tradeoff in executable world models, since improving observation fidelity can come at the expense of action-discriminative dynamics, and vice versa. Code is available at https://github.com/HKBU-KnowComp/PatchWorld.

21.
arXiv (CS.AI) 2026-06-15

When and How Severely: Scenario-Specific Safety Envelopes for Driving VLAs

arXiv:2606.14238v1 Announce Type: cross Abstract: Safety certification of Vision-Language-Action (VLA) driving planners under ISO 21448 (SOTIF) rests on an Operational Design Domain (ODD) specification that answers two complementary questions: when does the planner start to fail, and how severely does it fail once it does? We evaluate Alpamayo R1, a 10B-parameter open-weight driving VLA, on 15,968 (clip, attack) pairs. We find a conservative-aggregate gap: an aggregate safe threshold of $\sigma \leq 50$ under a 15% average displacement error (ADE) budget masks well-sampled scenarios that tolerate the top of the tested grid ($\sigma = 70$). A Gaussian Mixture Model (GMM) on the changed-explanation subset identifies six discrete severity bands (BIC-optimal $k{=}6$), so two perturbation conditions with the same mean error can differ materially in their share of high-severity (C4/C5) failures. Joining the two analyses on the same corpus surfaces a finding neither yields in isolation: the scenarios with the loosest noise thresholds are not those with the lowest high-severity rate: STOP_SIGNAL concentrates roughly $4\times$ the C4/C5 share of LANE_KEEPING despite tolerating a larger $\sigma$. A deployable SOTIF ODD specification for driving VLAs therefore requires a two-dimensional safety envelope, not a single aggregate value per hazard.

22.
arXiv (CS.LG) 2026-06-15

PERRY: Policy Evaluation with Confidence Intervals using Auxiliary Data

arXiv:2507.20068v2 Announce Type: replace Abstract: Off-policy evaluation (OPE) methods estimate the value of a new reinforcement learning (RL) policy prior to deployment. Recent advances have shown that leveraging auxiliary datasets, such as those synthesized by generative models, can improve the accuracy of OPE methods. Unfortunately, such auxiliary datasets may also be biased, and existing methods for using data augmentation within OPE lack principled uncertainty quantification. In high stakes domains like healthcare, reliable uncertainty estimates are important for ensuring safe and informed deployment of RL policies. In this work, we propose two methods to construct valid confidence intervals for OPE with data augmentation. The first provides a confidence interval over $V^{\pi}(s)$, the policy value conditioned on an initial state $s$. To do so we introduce a new conformal prediction method suitable for Markov Decision Processes (MDPs) with continuous state spaces, extending prior work to higher-dimensional settings. Second, we consider the more common task of estimating the average policy performance over all initial states, $V^{\pi}$; we introduce a method that draws on ideas from doubly robust estimation and prediction powered inference. Across simulators spanning inventory management, robotics, healthcare, and a real healthcare dataset from MIMIC-IV, we find that our methods can effectively leverage auxiliary data and consistently produce confidence intervals that cover the ground truth policy values, unlike previously proposed methods. Our work enables a future in which OPE can provide rigorous uncertainty estimates for high-stakes domains.

23.
arXiv (CS.CL) 2026-06-16

PreLort: Prefix-Nested LoRA for Federated Fine-Tuning under Rank Heterogeneity

Federated fine-tuning of large language models using parameter-efficient methods such as LoRA enables privacy-preserving adaptation of foundation models. Heterogeneous hardware resources introduce challenges, as clients with different adapter ranks cannot be directly aggregated. While existing methods enable aggregation under heterogeneous ranks, they fail to control how information is distributed across rank dimensions, leading to suboptimal use of shared low-rank representations. Instead, we propose PreLort: a nested low-rank formulation for federated LoRA that organizes adapter dimensions into a prefix hierarchy. Our approach ensures that lower-rank dimensions encode task-relevant information, while higher-rank dimensions capture additional capacity. Building on this, we introduce (i) a segment-wise aggregation rule that averages only over clients contributing to each rank segment, avoiding dilution from zero-padded lower-rank clients, and (ii) a prefix-nested training strategy that optimizes each adapter under multiple rank truncations, encouraging useful signal to concentrate in low-rank prefix dimensions. Together, these components encourage a consistent low-rank prefix capturing the most task-relevant information, while higher-rank dimensions learn additional capacity. This allows low-rank clients to benefit from richer information contributed by higher-rank clients, as prefix dimensions are consistently learned and aggregated. Experiments demonstrate that our method consistently outperforms prior heterogeneous federated LoRA methods in accuracy and ROUGE-L, while achieving lower or comparable perplexity across multiple base models.

24.
arXiv (CS.CL) 2026-06-18

STARE: Surprisal-Guided Token-Level Advantage Reweighting for Policy Entropy Stability

Reinforcement Learning with Verifiable Rewards algorithms like GRPO have emerged as the dominant post-training paradigm for complex reasoning in LLMs, yet commonly suffer from policy entropy collapse during training. We conduct a first-order gradient analysis of token-level entropy dynamics under GRPO and identify a token-level credit assignment mismatch: the per-token entropy variation decomposes into the product of the trajectory-level advantage and an entropy sensitivity function over the next-token distribution, yielding an advantage-surprisal four-quadrant structure and a near-criticality property. Motivated by it, we propose STARE (Surprisal-guided Token-level Advantage Reweighting for policy Entropy stability), which identifies entropy-critical token subsets via batch-internal surprisal quantiles, selectively reweights their effective advantages, and incorporates a target-entropy closed-loop gate for stable entropy regulation. Across model scales from 1.5B to 32B and three task families (Short CoT, Long CoT, and Multi-Turn Tool Use), STARE sustains stable RL training over thousands of steps while maintaining policy entropy within the target band. On AIME24 and AIME25, STARE outperforms DAPO and other competitive baselines by 4%-8% in average accuracy, with reflection tokens and response length growing in tandem, indicating sustained exploration-exploitation balance that further unlocks RL training potential.Code is available at https://github.com/hp-luo/STARE.

25.
arXiv (CS.CL) 2026-06-12

From Benchmarks to Skills: Low-Rank Factors for LLM Evaluation

Current evaluations of large language models (LLMs) rely heavily on a growing collection of benchmarks and on aggregate benchmark scores, yet it remains unclear what this comparison actually captures, and what these scores reveal about models' underlying capabilities. Here, we propose a new paradigm for LLM evaluation, by asking whether benchmark performance reflects many independent abilities, or rather relies on a small number of shared dimensions. To answer this, we apply Factor Analysis (FA) to a massive performance matrix of LLMs versus benchmarks \((60\times44)\) revealing an intrinsically low-rank structure of that matrix. That is, a small number of latent factors captures most of the structure in the full task space. This low-rank geometry reveals substantial redundancy across existing tasks and explains why many benchmarks appear to be measuring overlapping abilities. We further show that these latent factors correspond to coherent, skill-like, dimensions of LLM behavior. Leveraging this latent skill-space, we deliver three practical tools for LLM evaluation and downstream users: (i)~identifying redundant tasks, (ii)~profiling new models using a small subset of tasks, and (iii)~selecting models aligned with desired skill profiles. Our method provides a solid alternative to the de-facto standard of a single aggregate score, and establishes an interpretable and practical framework for understanding and benchmarking LLM core capabilities.