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) 2026-06-19

Impossibility of superluminal signalling rules out causal loops in conical spacetimes

arXiv:2606.20476v1 Announce Type: cross Abstract: In PRL 129, 110401 it was shown that it is theoretically possible to have operationally detectable causal loops without violating the principle of no superluminal signalling (NSS) in (1+1)-Minkowski spacetime. Whether or not such causal loops are also possible in $d > 1$ spatial dimensions, has remained a key open question. We resolve this question by showing that in a wide class of "conical" spacetimes, including Minkowski with d > 1, NSS does rule out all operationally detectable causal loops, in classical, quantum and post-quantum theories. This establishes that the relationship between the relativistic principles of NSS and no causal loops depends inherently on the geometry of spacetime.

02.
arXiv (CS.LG) 2026-06-17

When the Next Step Is Not One Step: Distribution-Aware Execution Modeling for Concurrent Go Programs

arXiv:2606.17508v1 Announce Type: new Abstract: Training a model to predict the next step in a concurrent program is harder than it looks: two runs of the same program from the same trace prefix can produce different next events, both valid, because the scheduler is nondeterministic. A model trained against a single label is learning to guess one outcome of a random process. We turn this around and use the nondeterminism as a training signal. We run each program many times, aggregate the observed next events into an empirical distribution, and fine-tune a 7B model to match that distribution with a KL objective. On 798 held-out predictions drawn from real production Go bugs (CockroachDB, Kubernetes, gRPC, etcd), fine-tuning on fewer than a thousand traces reaches 36.2% accuracy, ahead of Gemini 3.5 Flash used zero-shot (34.8%) and the same model without fine-tuning (28.6%). Distribution training matches cross-entropy on accuracy (35.8% vs. 36.2%) while reducing Expected Calibration Error from 0.205 to 0.169. We also derive a formal goroutine-leak signature for a class of select-blocked goroutines where P(GoUnblock)=0 holds by scheduler semantics, not by learning. We release the dataset, trained adapters, and all tooling.

03.
arXiv (CS.CL) 2026-06-19

NAMESAKES: Probing Identity Memorization in Text-to-Image Models

Text-to-image (T2I) models generate realistic likenesses of some individuals when prompted with their names, raising privacy concerns. However, distinguishing whether a generated face is memorized or fabricated currently requires ground-truth photos, access to training data, or white-box access to model internals, limiting applicability. We introduce a fully black-box behavioral probe that distinguishes between these regimes while requiring no reference photos or prior knowledge of training data. To benchmark this task, we present the NAMESAKES dataset of over one thousand names and faces of public figures spanning a wide range of fame levels, along with perturbed, less famous names. Experiments on state-of-the-art T2I models show that our probe substantially predicts identity memorization and separates memorized from unrecognized names, with further insights into differences across model families.

04.
medRxiv (Medicine) 2026-06-23

Socioeconomic Determinants of Guideline-Concordant Therapy for Early-Stage Non-Small Cell Lung Cancer: A Population-Based Analysis from Appalachian and Non-Appalachian Ohio, 2004-2015

Purpose: To examine the relative contributions of insurance, county-level poverty, and other socioeconomic factors, as compared with Appalachian geography, to receipt of guideline-concordant therapy for early-stage non-small cell lung cancer (NSCLC) in Appalachian and non-Appalachian Ohio. Methods: Retrospective population-based cohort study using the Ohio Cancer Incidence Surveillance System. We identified adults diagnosed with early-stage NSCLC between 2004 and 2015 (N=26,756). The primary outcome was receipt of guideline-concordant local therapy (surgery or definitive radiation). Rural-urban classification used USDA Rural-Urban Continuum Codes. Multivariable logistic regression and Cox proportional hazards models assessed predictors of treatment and survival, with E-values, race-stratified models, and propensity score weighting as sensitivity analyses. Findings: Median age was 71 years; 50.3% were male, 83.8% non-Hispanic White, and 20.4% Appalachian. Overall, 83.6% received guideline-concordant local therapy (59.6% surgery, 24.0% radiation). In adjusted analysis, Medicaid (adjusted odds ratio [OR] 0.53, 95% confidence interval [CI] 0.44-0.63; adjusted risk ratio [RR] 0.94, 0.91-0.96), county-level poverty >20% (OR 0.77, 95% CI 0.68-0.87; RR 0.96, 0.95-0.98), and unmarried status were independently associated with lower therapy receipt, whereas Appalachian residence was associated with modestly higher receipt (OR 1.17, 95% CI 1.06-1.29; RR 1.02, 1.01-1.04). Therapy rates converged across regions over the study period (year x Appalachian interaction p20% (HR 1.13, 95% CI 1.07-1.20). Conclusions: Socioeconomic factors, particularly Medicaid insurance and county-level poverty, were the patient characteristics most strongly associated with lower receipt of guideline-concordant therapy, whereas Appalachian residence was not a barrier. Findings support targeted interventions addressing insurance-related and poverty-related barriers to lung cancer care in high-poverty communities regardless of geographic designation.

05.
arXiv (quant-ph) 2026-06-16

Optimizing Wigner Negativity in Scattering Processes Using Energetic Cost Functions

arXiv:2606.15101v1 Announce Type: new Abstract: Wigner negativities (WNs) are key signatures of non-Gaussian bosonic states and essential resources for quantum technologies. We study their generation in the scattering of coherent pulses by a two-level atom coupled to a one-dimensional reservoir, a unitary and energy-preserving platform. Optimization in this multimode setting is hindered by the complexity of evaluating Wigner functions. We overcome this challenge by introducing energetic cost functions that identify output modes most likely to host large negativities. First using incoherent energy and then isolating a genuinely non-Gaussian contribution, we demonstrate a strong correlation between these quantities and WNs. This correlation extends beyond short, intense pulses to encompass pulses of finite energy, where photons are scattered while the two-level atom is driven. Focusing on the energy-efficiency of the process, we show that maximally efficient generation takes place for one input photon, on average, spectrally mode-matched with the atom.

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

Privacy from Symmetry: Orthogonally Equivariant Transformers for LLM Inference

arXiv:2606.16461v1 Announce Type: new Abstract: Running large language models locally is often impractical, pushing inference on sensitive text to third-party providers. Split inference partially mitigates this by keeping tokens on the client and sending only hidden representations, but these representations can still be recovered via nearest-neighbor search against the public embedding table. We propose an orthogonal obfuscation procedure in which the client multiplies embeddings by a secret orthogonal matrix before transmission. To enable correct inference under arbitrary rotations, we introduce ConjFormer, a transformer variant that is exactly $\mathrm{O}(d)$-equivariant via a lightweight normalization change (scalar RMSNorm) together with blockwise orthogonal conjugation of all linear weights. As a result, the server performs the full forward pass entirely in the rotated basis and never observes unrotated hidden states. Experiments on GPT-2 and Llama 3.2 1B models fine-tuned on PubMed show that orthogonal obfuscation eliminates direct cosine nearest-neighbor inversion and reduces token recovery from over 35% top-10 to at most 1.3%, while increasing perplexity by only 0.4% after fine-tuning. These results indicate that enforcing symmetry at the architectural level can provide a practical defense for privacy-preserving LLM inference without noise injection or heavy cryptographic machinery.

07.
arXiv (CS.CL) 2026-06-11

Quantifying Subliminal Behavioral Transfer Ratios in Language Model Distillation

Distillation of a language model intended to transfer benign behavior to a student model may also transfer undesirable characteristics, if they are present in the teacher model, a phenomenon known as subliminal learning. While qualitative evidence supports the existence of this effect, its magnitude has not been systematically characterized. This study quantifies subliminal behavioral transfer ratios by steering two teacher models (Llama-2-7B-Chat and Qwen2.5-7B-Instruct) at varying steering strengths and distilling student models using only benign data. Evaluation on 100 JailbreakBench prompts with GPT-4.1, serving as the evaluator, indicates that transfer is robust but exhibits distinct scaling behaviors. Llama-2 demonstrates a sharp threshold ($\tau = {0.25,0.32} \ beyond \ \alpha = -0.15$), whereas Qwen2.5 displays continuous and higher levels of transfer ($\tau$ up to $0.61$).

08.
arXiv (CS.AI) 2026-06-11

Continual Quadruped Robots Coordination via Semantic Skill Discovery

arXiv:2606.08102v2 Announce Type: replace-cross Abstract: Multi-quadruped coordination has attracted increasing attention due to its enhanced payload capacity, broader contact coverage, and improved adaptability to challenging tasks. Existing methods for multi-quadruped manipulation typically focus on predefined or closed task families, often relying on multi-agent reinforcement learning (MARL) to train task-specific coordination policies. However, such methods struggle in open-ended continual learning settings, where tasks arrive sequentially and robots are expected to acquire new coordination skills while reusing previously learned ones without catastrophic forgetting. To address this challenge, we propose Conquer, a semantic skill-library framework that formulates continual multi-quadruped coordination as a retrieve-adapt-update process. First, to accommodate varying team sizes across tasks, we design a team-structured Self-Allies-Goal (SAG) backbone that supports variable-cardinality robot teams by explicitly modeling each robot's own state, teammate context, and task goal. For each incoming task, Conquer constructs a task-level semantic descriptor from pre-execution information and retrieves a relevant skill from the library for adaptation. After successful execution, Conquer updates the skill library by extracting trajectory-level semantic descriptors and organizing them according to semantic distance, thereby enabling continual skill accumulation and cross-task knowledge transfer. Simulation experiments show that Conquer achieves a final average success rate of 95.6%, demonstrating strong forward transfer and negligible catastrophic forgetting. Real-world rollouts on Unitree Go2 teams further validate the deployment feasibility of Conquer for practical multi-quadruped coordination. Simulation and real-robot demonstration videos are available at: https://conquer-project.pages.dev/.

09.
arXiv (CS.AI) 2026-06-16

A Security Analysis of Long-Horizon Agentic AI Systems: Threats, Evaluation, and Framework Development

arXiv:2606.14816v1 Announce Type: cross Abstract: This paper presents a structured analysis of security challenges in long-horizon agentic AI systems. The study reviews existing threats, evaluation approaches, attack propagation mechanisms, and security frameworks. A taxonomy of security threats and a framework for analyzing attack propagation are proposed to support future research in agentic AI security

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

Boosting Direct Preference Optimization with Penalization

作者:

arXiv:2606.12505v1 Announce Type: cross Abstract: Offline preference optimization has become a practical substitute for reinforcement learning from human feedback, but pairwise objectives such as Direct Preference Optimization (DPO) and its variants use only the chosen and rejected responses stored in a static dataset. This leaves a useful signal unused: the response that the reference model itself would generate for the same prompt. We propose Direct Preference Optimization with Penalization (DPOP), a simple extension of DPO that augments the base preference loss with a gated penalty on reference-greedy responses. DPOP activates this penalty only when the current policy still assigns a lower likelihood to the preferred response than to the rejected response. On AlpacaEval 2.0, DPOP improves length-controlled win rate over DPO, SimPO, and AlphaDPO on both Llama-3-8b-it and Gemma-2-9b-it, achieving relative gains of 5.3\% and 4.4\% over baselines on the two models, respectively. Ablations further show that a SimNPO-style length-normalized penalty is stronger than NPO and token-level unlikelihood in this setting.

11.
medRxiv (Medicine) 2026-06-16

Presurgical immune biomarkers associated with pain intensity and pain interference recovery after total knee arthroplasty: findings from the PRIME-KNEE study

Chronic postsurgical pain (CPSP) prevalence after total knee arthroplasty (TKA) is >20%. Circulating immune biomarkers are known factors of musculoskeletal pain but poorly understood as CPSP predictors. This prospective, longitudinal study of 203 patients s/p TKA tested presurgical plasma biomarkers associated with 6-month CPSP, using promising approaches from geriatrics biomarker research: expected recovery differential (ERD; resilience outcome) and penalized, machine-learning regularization modeling (elastic net and LASSO regression). Forty-nine presurgical candidate biomarkers were considered. CPSP was operationalized using ERDs built around PROMIS pain intensity and pain interference, which quantified the difference between observed and expected recovery after accounting for demographic, comorbidity, reserve, and perioperative factors. Plasma/ERDs from ~130 patients revealed 13 biomarkers with the highest selection stability criteria, and either positive or negative (+/-) associations with ERDs. Interleukin (IL) 5 (-) and Lipopolysaccharide-Binding Protein (LBP; +) were associated with both ERDs. Unique associations with pain intensity ERD included Cytomegalovirus-Specific IgG Negative (CMV IGg-; -), Macrophage Inflammatory Protein-1 Beta (MIP1b; -), IL12p70 (-, Cluster of Differentiation 30 (sCD30;-), Interferon alpha 2a (IFN2a;+), and Leukemia Inhibitory Factor (LIF;+). Unique associations with pain interference ERD included Lipopolysaccharide (LPS;-), Activin A (-), IL8 (-), Serum Amyloid A (SAA;-), and IL7 (+). Protein-protein interaction analyses and topology motifs suggest a centralized network with higher-than-expected connectivity, involving IL5, IL7, IL8, MIP1{beta}, and IFN2a, among others. This study proposes rigorous yet feasible approaches to expedite pain biomarker research, and introduces presurgical biomarkers t0 consider in future TKA-CPSP biosignature derivation.

12.
arXiv (CS.LG) 2026-06-16

SPICE: Synergy and Partial Information Based Curriculum Evolution

arXiv:2606.16639v1 Announce Type: new Abstract: Multimodal learning exploits complementary information across heterogeneous modalities. The informativeness of each modality can vary widely across samples and training stages. Existing multimodal curriculum learning strategies often assume that the relative complexity of samples remains unchanged throughout training and therefore cannot adapt to model evolution. We propose SPICE (Synergy and Partial Information based Curriculum Evolution), a novel progressive curriculum framework for multimodal interaction learning. Guided by Partial Information Decomposition (PID) theory, our approach decomposes multimodal interactions into redundant, unique, and synergistic information components, enabling an interpretable and dynamic characterization of sample complexity. Building on this decomposition, we design a progressive curriculum that evolves throughout training, allowing the model to transition from learning shared cross-modal cues to modality-specific patterns and, finally, to complex synergistic interactions. Adapting to model evolution, sample ordering is refined in real-time using PID information estimates derived from unimodal and multimodal predictions. Experiments across multiple multimodal benchmarks demonstrate consistent improvements over conventional training and state-of-the-art baselines, highlighting the effectiveness of PID information decomposition and adaptive sample ordering for multimodal curriculum learning.

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

Is Your Trajectory Displacement Safe in Long-tail?

arXiv:2606.16313v1 Announce Type: cross Abstract: Long-tail scenarios remain a major bottleneck for autonomous driving evaluation, even as datasets grow by orders of magnitude. Existing evaluation pipelines are rarely human-aligned, safety-aware, verifiable, and explainable at the same time: closed-loop metrics often saturate among strong planners, while unstructured human ratings can be noisy without a carefully designed protocol. We formulate planning evaluation as additional-threat detection: given a planner trajectory and an expert reference, does the planner's displacement introduce new unsafe driving behavior? We propose FluidTest, an evaluation pipeline with three components: a pairwise WebUI protocol for reliable human annotation; a taxonomy of 32 semantic threats with evidence-grounded decision graphs; and a three-agent verification system with reflection for precision and auditability. Experiments on the WOD-E2E dataset show that FluidTest produces consistent labels among trained annotators and identifies additional threats in 65% of Poutine trajectories and 51% of RAP trajectories. These results show that state-of-the-art planners can still exhibit substantial safety-relevant failures despite high Rater Feedback Scores (RFS) and low Average Displacement Error (ADE). Additional details, guidance, and code are available at https://fluidtest.web.app.

14.
arXiv (CS.CV) 2026-06-19

Neural Events: Discrete Asynchronous Autoencoders for Event-Based Vision

Event cameras capture dynamic scenes with exceptional temporal fidelity by representing them as a continuous stream of microsecond resolution events. Each individual event, however, only carries minimal semantic value, merely signaling a localized brightness change. To derive meaningful signals, downstream algorithms need to quickly integrate cues from a potentially massive torrent of low-information events. Current architectures, however, are easily overwhelmed, struggling to balance capturing fine-grained temporal dynamics and maintaining a manageable data throughput. This paper proposes a framework to re-tokenize event streams into a small set of highly informative neural events, each representing a local spatio-temporal context window with a discrete learnable code. Every time this code flips, a neural event is triggered, yielding a highly compressed data stream. We demonstrate that, across object detection and classification, networks trained on neural events are on par or surpass the performance of state-of-the-art approaches while reducing the event rate by a factor of 2.0.

15.
arXiv (CS.CV) 2026-06-18

A Survey on Deep Learning Architectures for Point Cloud Classification and Segmentation

Point cloud stands as the most widely adopted format for representing 3D shapes and scenes due to its simplicity and geometric fidelity. However, its inherent unordered and irregular nature, exacerbated by sensor noise and occlusions, introduces unique challenges for machine learning based methodologies. To combat these issues, diverse strategies have been developed, including converting to a format that has orderliness, extracting local geometry, and permutation-invariant or self-attention-based processing. In this paper, our focus is directed towards deep learning models for three fundamental tasks in 3D vision: point cloud classification, part segmentation, and semantic segmentation. We begin by formally defining point cloud data, followed by an in-depth discussion on its structural characteristics. Then, we categorize notable works based on their backbone structure and evaluate their performance on popular benchmarks. Beyond empirical comparison, we offer insights into architectural innovations and limitations. We also outline open challenges and promising future directions for 3D point cloud understanding.

16.
medRxiv (Medicine) 2026-06-10

Developing a Unified Criminal Justice Pathway into Drug and Alcohol Treatment from Police Custody: A Public Health Service Evaluation and Pathway-Design Project in Blackpool, United Kingdom

Introduction: Blackpool, England's most deprived local authority, has the highest drug-related death rate in the country. People in police custody with problem substance use are a key Core20PLUS5 inclusion-health group, yet referral from the police into structured drug and alcohol treatment is fragmented and relies heavily on self-report. We evaluated the current police-to-treatment route in Blackpool and designed an evidence-informed unified pathway. Materials and Methods: A mixed-methods service evaluation and pathway-design project was conducted during a six-month General Practice / Public Health rotation. Routinely collected referral data from Horizon (the local specialist drug and alcohol service) covering the 47-month period from December 2019 to October 2023 were analysed. Findings were triangulated with national policy, the Project ADDER and Liaison and Diversion evaluations, and the international evidence on police-led pre-arrest diversion. Results: Of 5,900 total referrals into Horizon over 47 months, only 269 (4.56%) originated from the police. Police referrals accounted for fewer than 5% of monthly referrals in 30 of 47 months, for 5 to 9.9% in 16 months, and for >/= 10% in only one month (10.8%, December 2022). Blackpool recorded 76 drug-misuse deaths in 2019-21 (19.4 per 100,000, approximately four times the England rate). A six-step unified pathway is proposed: Initiate Referral (opt-out, from ADDER Police and Liaison and Diversion); Initial Assessment; Tailored Treatment Plan; Continuous Support; Collaboration and Monitoring; and Evaluation and Adjustment. Conclusions: Police contact is markedly under-used as a gateway to treatment despite Blackpool having the highest drug-related mortality in England. An opt-out, multi-agency pathway anchored in Core20PLUS5 has the potential to narrow the treatment gap, reduce re-offending, and address the structural health inequalities that drive premature mortality.

17.
arXiv (CS.CL) 2026-06-16

Dealing with Annotator Disagreement in Hate Speech Classification

Hate speech detection is a crucial task, especially on social media where harmful content can spread quickly. Collecting social media content (tweets etc.) to train machine learning models is easy, but detecting and categorizing hate speech can be difficult due to the inherently subjective nature. This subjectivity leads to frequent disagreement among annotators, particularly for subtle or borderline content. Traditional approaches either discard non-consensus samples or force a ''gold standard'' through expert adjudication, ignoring valuable information about uncertainty and diverse human perspectives. We examine the largely overlooked problem of annotator disagreement in hate speech classification and evaluate a range of aggregation methods, including majority voting, ordinal strategies (minimum, maximum, and mean), and analyze their impact across binary, 4-class, and 6-class classification tasks. In addition, we leverage annotators' perceived hate speech strength scores to explore regression-based and hybrid modeling approaches. Among others, we show that filtering non-consensus samples results in over-optimistic results and that the perceived strength provides a complementary signal that enhance classification performance. Finally, we establish new state-of-the-art results for hate speech detection in Turkish tweets, and demonstrate that annotator disagreement, when properly modeled, is a valuable resource for building more robust and reliable systems.

18.
arXiv (CS.AI) 2026-06-19

Analyzing the Narration Gap in LLM-Solver Loops

arXiv:2606.19588v1 Announce Type: new Abstract: Formal tools such as SAT and SMT solvers are increasingly embedded in language model reasoning pipelines when a safety or security critical question can be formulated in logic. Unlike chain of thought whose steps are sampled from the model distribution without formal guarantee, a solver produces a sound and independently verifiable answer. However, the soundness guarantee can be lost in the interaction between the solver and the model. The hybrid pipeline has three components: formalizing the question, deciding it, and narrating the result. Prior work has studied the formalization and decision, but not narration, which is the step that turns a formal tool's output into the user answer. To fill the narration gap, we first model the LLM-solver loop as a verified decision procedure. We further evaluate five open-sourced models under prompt injection, and we find certificate gating makes the solver verdict sound, while an adversary can invert a verified conclusion across phrasings and channels. We study the mitigation through hardened prompt that reduces injection significantly but cannot eliminate it and still suffers under adaptive attack. Combining the formal analysis and empirical studies, we show in the LLM-solver loop, robustness does not reach to the answer that the user finally reads.

19.
arXiv (CS.CL) 2026-06-16

Control-Plane Placement Shapes Forgetting: An Architectural Study of Agent Memory Across Thirteen System Configurations

作者:

Where an LLM sits in an agent memory pipeline – between the recall plane that retrieves stored facts (extensively benchmarked) and the control plane that mutates them via supersede, release, purge (largely untested) – shapes which forgetting failure modes the system recovers. Comparing thirteen system configurations on a 385-case adversarial surface, we observe three placement regimes with partly complementary coverage: deterministic primitives suffice for lexical/temporal categories but fail canonicalization (5% on identifier-obfuscation, 0% on cross-lingual); inscribe-time LLM recovers canonicalization (100%) but cannot help intent-aware deletion (0% on prefix-collision and compound-fact); a mutation-time hook recovers intent-aware deletion (78-85%) and brightens nearly all categories simultaneously (91.7-93.2% overall, $0.17 per 385-case run, 2.3s/case mutation latency vs. 64-191ms/case deterministic, recall path unchanged). We expose the trade-off via ForgetEval, a 1000-case templated suite plus a 385-case adversarial layer (132 hand-crafted + 253 LLM-drafted oracle-validated) scored by deterministic substring match, paired with a six-method Adapter Protocol with honest N/A scoring that lets heterogeneous memory stores enter in 130 lines. Admission is corroborated by 10-annotator IAA (Fleiss' kappa = 0.958) and a 77-case external-authored subset (four blind contributors) that replicates the canonicalization asymmetry and amplifies the joint-placement lift (+27.8 pt). Production failures are predominantly forgetting failures rather than recall failures, yet existing benchmarks measure only recall. ForgetEval and all adapters are released under MIT.

20.
arXiv (CS.AI) 2026-06-12

Competition and Diversity in Generative AI

arXiv:2412.08610v3 Announce Type: replace-cross Abstract: Recent evidence, both in the lab and in the wild, suggests that the use of generative artificial intelligence reduces the diversity of content produced. The use of the same or similar AI models appears to lead to more homogeneous behavior. Our work begins with the observation that there is a force pushing in the opposite direction: competition. When producers compete with one another (e.g., for customers or attention), they are incentivized to create novel or unique content. We explore the impact competition has on both content diversity and overall social welfare. Through a formal game-theoretic model, we show that competitive markets select for diverse AI models, mitigating monoculture. We further show that a generative AI model that performs well in isolation (i.e., according to a benchmark) may fail to provide value in a competitive market. Our results highlight the importance of evaluating generative AI models across the breadth of their output distributions, particularly when they will be deployed in competitive environments. We validate our results empirically by using language models to play Scattergories, a word game in which players are rewarded for answers that are both correct and unique. Overall, our results suggest that homogenization due to generative AI is unlikely to persist in competitive markets, and instead, competition in downstream markets may drive diversification in AI model development.

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

Beyond Rubrics: Exploration-Guided Evaluation Skills for Reward Modeling

Open-ended reward modeling requires judges that can follow subtle, domain-specific preferences when verifiable answers are unavailable. Existing rubric-based methods often address this by generating criteria online for each query, but the extra generation step can add inference overhead and produce rigid or misaligned guidance. We introduce Eval-Skill, an exploration-guided method that synthesizes reusable evaluation skills for reward modeling and reframes reward guidance as context evolution rather than parameter training or per-query rubric generation. Using only 100 cases per domain for skill evolution, Eval-Skill synthesizes reusable domain-level evaluation skills through two progressive stages, workflow generation followed by principle generation, with exploration and selection interleaved across both stages. Once generated, a skill is directly injected into the judge context. Across multiple RM benchmarks, Eval-Skill consistently improves diverse judge backbones; on RewardBench 2, it yields significant gains over vanilla judging for each main backbone (+13.44% for Qwen3-8B, and 18.51% for DeepSeek-V4-Flash). Further analyses of evolution-time scaling, generalizability, and transferability show that compact evaluation skills offer an efficient new paradigm for LLM-based evaluation. Code is available at https://github.com/xing-stellus-yue/Eval-Skill.

22.
arXiv (CS.CL) 2026-06-17

Prompt Perturbation for Reliable LLM Evaluation over Comparison Graphs

Evaluating large language models (LLMs) is important for understanding their capabilities, comparing competing systems, and supporting the deployment of reliable models in practice. For open-ended tasks, pairwise evaluation has become a popular paradigm, in which two responses to the same prompt are compared and the resulting judgments are aggregated into an overall ranking. A central challenge of this paradigm is intransitivity: the induced comparison outcomes may fail to support any coherent global ranking. For example, one may observe cyclic preferences such as $A \succ B \succ C \succ A$, or inconsistencies involving ties such as $A \equiv B\equiv C\neq A$. Such contradictions make the resulting leaderboard unstable and challenging to interpret. In this paper, we propose a prompt perturbation framework for improving the consistency of pairwise LLM evaluation. Our approach generates perturbed variants of each prompt, uses the resulting comparison graphs to identify and filter out structurally inconsistent comparison patterns, and then applies standard ranking methods to the filtered comparisons. A key feature of the proposed framework is that graph-level structural consistency is incorporated explicitly into the evaluation pipeline before ranking aggregation. This provides a simple and principled way to reduce cyclic inconsistencies and improve the reliability of LLM rankings.

23.
arXiv (math.PR) 2026-06-15

Real-order moments, tail representations, and logarithmic means

arXiv:2606.14019v1 Announce Type: cross Abstract: This paper develops a unified framework for the study of real-order moments of arbitrary random variables. General integral representations are established in terms of cumulative distribution functions and survival functions, covering continuous, discrete, and mixed distributions supported on the whole real line. These formulas extend the classical tail-integral identities for nonnegative random variables and provide a common treatment of positive, fractional, and negative moments. For discrete distributions, explicit series representations are derived in terms of cumulative probabilities, yielding simple criteria for the existence of moments. Applications are presented for the zeta and Skellam distributions, illustrating how tail behavior determines moment finiteness and how moments can be represented geometrically through cumulative distribution functions. In addition, a representation for logarithmic moments is obtained, linking logarithmic means, Laplace transforms, and the classical Frullani identity. The results provide a unified perspective on moment representations and establish useful connections between tail probabilities, distribution functions, Laplace transforms, and moment existence.

24.
arXiv (CS.CL) 2026-06-15

Rethinking the Trust Region in LLM Reinforcement Learning

Reinforcement learning (RL) has become a cornerstone for fine-tuning Large Language Models (LLMs), with Proximal Policy Optimization (PPO) serving as the de facto standard algorithm. Despite its ubiquity, we argue that the core ratio clipping mechanism in PPO is structurally ill-suited for the large vocabularies inherent to LLMs. PPO constrains policy updates based on the probability ratio of sampled tokens, which serves as a noisy single-sample Monte Carlo estimate of the true policy divergence. This creates a sub-optimal learning dynamic: updates to low-probability tokens are aggressively over-penalized, while potentially catastrophic shifts in high-probability tokens are under-constrained, leading to training inefficiency and instability. To address this, we propose Divergence Proximal Policy Optimization (DPPO), which substitutes heuristic clipping with a more principled constraint based on a direct estimate of policy divergence (e.g., Total Variation or KL). To avoid huge memory footprint, we introduce the efficient Binary and Top-K approximations to capture the essential divergence with negligible overhead. Extensive empirical evaluations demonstrate that DPPO achieves superior training stability and efficiency compared to existing methods, offering a more robust foundation for RL-based LLM fine-tuning. Our code is available at https://github.com/sail-sg/Stable-RL.

25.
arXiv (CS.LG) 2026-06-15

MAD: Manifold Attracted Diffusion

arXiv:2509.24710v3 Announce Type: replace-cross Abstract: Score-based diffusion models are a highly effective method for generating samples from a distribution of images. We consider scenarios where the training data comes from a noisy version of the target distribution, and present an efficiently implementable modification of the inference procedure to generate noiseless samples. Our approach is motivated by the manifold hypothesis, according to which meaningful data is concentrated around some low-dimensional manifold of a high-dimensional ambient space. The central idea is that noise manifests as low magnitude variation in off-manifold directions in contrast to the relevant variation of the desired distribution which is mostly confined to on-manifold directions. We introduce the notion of an extended score and show that, in a simplified setting, it can be used to reduce small variations to zero, while leaving large variations mostly unchanged. We describe how its approximation can be computed efficiently from an approximation to the standard score and demonstrate its efficacy on toy problems, synthetic data, and real data.