探索全球前沿学术脉络

01.

arXiv (CS.AI) 2026-06-12 DOI: arXiv:2606.13097

Functional Cache Grafting: Robust and Rapid Code-Policy Synthesis for Embodied Agents

作者:

Saehun Chun↗Wonje Choi↗Sera Choi↗Sanghyun Ahn↗Honguk Woo↗

arXiv:2606.13097v1 Announce Type: cross Abstract: Code-writing large language models (CodeLLMs) generate executable code policies for embodied agents by translating natural language goals and environmental constraints into structured control programs. However, policy generation in open-domain embodied environments suffers from two fundamental limitations: (i) delayed decoding caused by repetitive prefill computation over long prompts, and (ii) limited robustness due to fully generative decoding, which often produces API mismatches, missing safety guards, and unstable control logic. To address these limitations, we present FCGraft, a Functional Cache Grafting framework. FCGraft maintains a library of function-level validated code skeletons and their associated prompt-level Transformer key-value (KV) caches, and synthesizes new policies by retrieving relevant functions and grafting their KV caches when a new task is provided. Given retrieved function caches, FCGraft performs cache grafting via stitching, which composes cached function segments into a composite policy, and patching, which locally adapts only the necessary code regions to satisfy task-specific parameters and constraints with minimal additional decoding. By eliminating redundant prefill computation, this approach reduces generation latency, while reusing validated control structures improves robustness over prompt-level caching methods RAGCache, achieving 18.31% higher task success rate and 2.3x faster policy synthesis.

阅读与讨论 → 访问原文 →

02.

arXiv (CS.LG) 2026-06-17 DOI: arXiv:2606.17603

Expanding SPHERE-JEPA: A Family of Statistical Regularizers for the Hypersphere

作者:

L\'eo Nicollier↗Enric Meinhardt-Llopis↗Max Dunitz↗Marc Pic↗Pablo Mus\'e↗Gabriele Facciolo↗

arXiv:2606.17603v1 Announce Type: new Abstract: In Self-Supervised Learning (SSL), preventing representation collapse by explicitly enforcing a uniform distribution on the unit hypersphere has proven to be effective. However, current frameworks typically rely on sliced statistical regularizers such as SIGReg (used in LeJEPA) and SUSReg (used in SPHERE-JEPA), which approximate this continuous objective via Monte Carlo sampling along random 1D directions. This stochasticity injects projection variance into the training gradients, destabilizing optimization, and hindering convergence. In this work, we first show that analytically integrating out these random projections natively yields a deterministic Maximum Mean Discrepancy (MMD), bypassing the variance of sliced methods. Motivated by this equivalence, we formulate full-dimensional objectives for MMD, Kernel Stein Discrepancy (KSD), and Kullback-Leibler (KL) divergence directly on the sphere to enforce a uniform distribution. To prevent spatial bias, we equip these tests with rotationally invariant kernels constructed via spectral theory, systematically evaluating two canonical families: smooth exponential decay (Heat) and strict frequency cutoff (Bandlimited) filters. Empirically, removing projection-induced noise results in more stable optimization, faster convergence, and consistent improvements over stochastic sliced regularizers on ImageNet and Galaxy10. Furthermore, we reveal that the choice of the statistical test shapes the geometry of the learned latent space: MMD and KSD favor locally clustered organization suitable for object-centric domains, whereas the continuous KDE-based KL divergence promotes fine-grained instance separation, yielding the strongest results on unclustered procedural texture retrieval.

阅读与讨论 → 访问原文 →

03.

arXiv (CS.AI) 2026-06-15 DOI: arXiv:2606.13705

Can Editing 1 Neuron Fix Repetition Loops in LLMs?

作者:

Aristotelis Lazaridis↗Aman Sharma↗Dylan Bates↗Brian King↗Vincent Lu↗Jack FitzGerald↗

arXiv:2606.13705v1 Announce Type: cross Abstract: Yes. Can it cure doom loops? Probably not. The Gemma 4 instruction-tuned models share a reproducible failure: on long factual enumeration prompts, such as listing every episode of a TV series, the 88 IAU constellations, or the 151 original Pokemon, they collapse into repetition, either a tight verbatim loop or a list whose entries decay onto a single answer. These loops occur at rates as high as 95% and survive prompt rewording, inference-engine changes, and most sampling adjustments. In this paper we explore whether this behavior is localized enough to remove by weight edits. To localize the cause, we use per-layer ablation and per-neuron attribution, then confirm the strongest candidates with full-generation sweeps. The loops trace to a small set of MLP neurons (or, in the 26B-A4B Mixture-of-Experts model, a few routed experts) which we suppress with static weight edits. These "surgeries" can be as small as a single sign-inverted neuron (in the E2B model). The size of the effective edits grows with model scale, but in all cases, the loop patterns can be addressed at normal generation budgets while preserving general-purpose benchmark scores. However, the edits do not solve everything: we also study longer thinking budgets, where the two larger models most visibly enter doom looping, i.e. a non-convergent regime in which the model self-corrects in circles over a fact it cannot recall, exhausting the budget without committing to a final answer. We show this residual failure is reduced but not eliminated by the same edits, and argue it is fundamentally a knowledge-precision problem rather than a removable circuit; weight surgery can delete a loop, but it cannot supply a missing fact. Our results are both a feasibility demonstration, that is, evidence that a concrete generation pathology can be localized to a few parameters and edited out, and a delineation of where that approach stops.

阅读与讨论 → 访问原文 →

04.

arXiv (CS.CV) 2026-06-11 DOI: arXiv:2606.08744

MB-Loc: Multi-planar Bird's-eye-view Localization in outdoor LiDAR scenes

作者:

Ayaan Choudhury↗Preet Savalia↗Anirudh Pydah↗Avinash Sharma↗

Global LiDAR localization is a fundamental task for autonomous navigation systems. Recent methods perform Scene Coordinate Regression (SCR) and achieve superior accuracy over Absolute Pose Regression (APR) solutions by predicting dense 3D world coordinates. However, SCR approaches introduce two major bottlenecks: severe computational inefficiency from processing raw 3D geometries and significant performance degradation under varying sensor viewpoints. To address these limitations, we present MB-Loc, a lightweight and viewpoint-robust SCR framework. Instead of relying on heavy 3D convolutions, we project the input LiDAR scan into a 2.5D Multi-planar Bird's-Eye View (BEV) representation. By slicing the point-cloud along the Z-axis and mapping signed depths into discrete 2D planes, MB-Loc retains essential 3D geometric structures while exploiting the computational tractability of standard 2D CNNs. To handle the inherent sparsity of outdoor LiDAR, we introduce a KL-regularized latent bottleneck that explicitly models spatial uncertainty without injecting stochastic noise. Finally, to ensure rotation robustness, we apply 3D spatial augmentations prior to planar projection, forcing the network to implicitly learn viewpoint-invariant features. We perform extensive experiments on the publicly available NCLT dataset and demonstrate that our proposed method outperforms the current state-of-the-art. Operating at real-time inference speeds, MB-Loc significantly outperforms traditional 3D-SCR architectures in computational efficiency.

阅读与讨论 → 访问原文 →

05.

arXiv (CS.LG) 2026-06-16 DOI: arXiv:2606.15386

A Compositional Framework for Open-ended Intelligence

作者:

Ida Momennejad↗Roberta Raileanu↗

arXiv:2606.15386v1 Announce Type: new Abstract: Open-ended intelligence is the capacity to adapt to novel problems and environments that are substantially different from those in training. We formalize open-ended intelligence as the closure induced by a finite primitive set \(P\) and a set of composition operators \(C\). We characterize properties of the induced closure \(\mathcal{L}(P,C)\) that support unbounded compositional generation across families of tasks and worlds. A mathematics of open-ended intelligence requires two pillars: a minimal set of representational primitives (e.g., states, actions) and algorithmic primitives (e.g., nearest neighbor), together with composition motifs (e.g., recursion, sequencing) that reflect an acquired compositional grammar. The closure of these two pillars enables the generation of infinite adaptive responses across a wide range of settings. The mathematics supports complementary research agendas, including evaluation metrics for explanation and interpretability, as well as building architectures where compositional generalization is native. We propose next primitive prediction as a novel architectural objective, where the training objective encourages the acquisition of reusable algorithmic primitives and their compositional grammar, such that new solutions are generated through recombination. Curriculum learning and self-play enable lifelong learning and expansion of the closure by discovering reusable primitives and transition motifs across families of tasks and worlds. We ground the framework through case studies in physics, evolution, and neuroscience.

阅读与讨论 → 访问原文 →

06.

arXiv (CS.CV) 2026-06-16 DOI: arXiv:2606.15967

CRIS: Cross-Plane Self-Supervised Isotropic Restoration for Anisotropic Volumetric Imaging Across Modalities

作者:

Adi Ahituv↗Anat Ilivitzki↗Moti Freiman↗

Anisotropic volumetric acquisitions are common in clinical MRI and volume electron microscopy (vEM), where sparse through-plane sampling creates thick slices or sections that degrade orthogonal reformats and downstream analysis. We present CRIS, a cross-plane self-supervised framework for isotropic restoration without paired isotropic ground truth. CRIS casts 3D restoration as 2D stripe completion on orthogonal reformats of an isotropic grid: high-resolution in-plane slices are synthetically degraded and periodically masked for training, while at inference blank slices define the isotropic grid, two orthogonal reformats are restored, and predictions are fused by multi-view averaging. We evaluate CRIS on two MRI cohorts and two microscopy benchmarks up to 8x anisotropy. On brain MRI, CRIS achieves 32.921 +/- 0.436 dB PSNR and 0.9631 +/- 0.0027 SSIM, outperforming interpolation, SMORE4, SIMPLE, SA-INR, and ATME, and gives the best segmentation consistency (Dice 0.940 +/- 0.004, ASSD 0.245 +/- 0.014 mm, HD99 1.275 +/- 0.061 mm). On reference-free abdominal MRI, CRIS reduces FID/KID to 48.714/0.023. On vEM, CRIS outperforms interpolation, NIIV, and vEMINR, reaching 29.133 dB/0.834 3D PSNR/SSIM at 4x, 27.123 dB/0.734 on EPFL at 8x, and 21.915 dB/0.699 on noisy hemibrain data. In a robustness experiment, one variable-gap CRIS model evaluated across gap factors 3–7 and coronal, axial, and sagittal degradations maintained higher PSNR/SSIM than interpolation (36.36–31.14 dB and 0.977–0.932 vs. 33.07–27.85 dB and 0.951–0.853). These results support CRIS as a modality-flexible route to isotropic restoration without paired isotropic targets or configuration-specific retraining. Code is available at https://github.com/adi-hatav/CRIS.

阅读与讨论 → 访问原文 →

07.

arXiv (CS.AI) 2026-06-12 DOI: arXiv:2606.12921

LoRA-Muon: Spectral Steepest Descent on the Low-Rank Manifold

作者:

Franz Louis Cesista↗Katherine Crowson↗C\'edric Simal↗Stella Biderman↗

arXiv:2606.12921v1 Announce Type: cross Abstract: Low-Rank Adaptation (LoRA) significantly reduces compute and memory costs for finetuning Deep Learning models but is often harder to tune than dense training: when using factor-wise optimizers such as AdamW, it is sensitive to initialization choices, its optimal learning rates transfer poorly across ranks, and it often fails to beat dense baselines. We derive LoRA-Muon by applying the Muon optimizer's spectral steepest-descent rule to the low-rank setting. Along with our split weight-decay rule, our main claim is that LoRA-Muon is a good low-rank proxy for full-rank Muon and Shampoo-family optimizers. Its optimal learning rates transfer across rank, width, depth, and factor-rescaling. In our compute-matched TinyShakespeare study, a rank-$2$ proxy recovers the dense best tested learning rate, and a rank-$32$ LoRA-Muon run attains lower mean validation loss than the dense baseline in the seed-averaged sweep. We further show that the Spectron optimizer depends on arbitrary factor scaling, so it would likely be a poor fit when finetuning starts from badly imbalanced factors, and that LoRA-RITE's simplified QR-coordinate core implements the same spectral update. LoRA-Muon computes that update without QR-decomposition and avoids storing second moments, making it more accelerator-friendly and memory-efficient.

阅读与讨论 → 访问原文 →

08.

arXiv (CS.AI) 2026-06-18 DOI: arXiv:2606.19145

OrthoReg: Orthogonal Regularization for Hybrid Symbolic-Neural Dynamical Systems

作者:

Till Richter↗Niki Kilbertus↗

arXiv:2606.19145v1 Announce Type: cross Abstract: Dynamical systems are fundamental to modeling the natural world, yet modeling them involves a persistent trade-off: manually prescribed mechanistic models are interpretable by design but often overly simplistic and misspecified; in contrast, flexible data-driven neural methods lack physical insight. Hybrid modeling aims for the best of both worlds by combining a prescribed or symbolic, physics-based component with a flexible neural network. A critical challenge, however, is that the neural component may relearn mechanistic parts, yielding redundant and uninterpretable models, especially when the symbolic structure itself is discovered from data. Existing methods based on standard $L^2$ regularization rely on a projection argument that breaks when the symbolic component is learned through sparse discovery, allowing the neural augmentation to overlap with symbolic structure. We introduce OrthoReg (Orthogonal Regularization), which directly penalizes overlap between the symbolic and neural components, preventing symbolic structure from being absorbed by the neural residual. This yields a complementary decomposition: the symbolic part captures what the library can express, and the neural part captures what remains. On benchmark dynamical systems with partial library mismatch, OrthoReg improves symbolic recovery and out-of-distribution behavior.

阅读与讨论 → 访问原文 →

09.

arXiv (CS.LG) 2026-06-16 DOI: arXiv:2602.24012

InfoNCE Induces Gaussian Distribution

作者:

Roy Betser↗Eyal Gofer↗Meir Yossef Levi↗Guy Gilboa↗

arXiv:2602.24012v2 Announce Type: replace Abstract: Contrastive learning has become a cornerstone of modern representation learning, allowing training with massive unlabeled data for both task-specific and general (foundation) models. A prototypical loss in contrastive training is InfoNCE and its variants. In this work, we show that the InfoNCE objective induces Gaussian structure in representations that emerge from contrastive training. We establish this result in two complementary regimes. First, we show that under certain alignment and concentration assumptions, projections of the high-dimensional representation asymptotically approach a multivariate Gaussian distribution. Next, under less strict assumptions, we show that adding a small asymptotically vanishing regularization term that promotes low feature norm and high feature entropy leads to similar asymptotic results. We support our analysis with experiments on synthetic and CIFAR-10 datasets across multiple encoder architectures and sizes, demonstrating consistent Gaussian behavior. This perspective provides a principled explanation for commonly observed Gaussianity in contrastive representations. The resulting Gaussian model enables principled analytical treatment of learned representations and is expected to support a wide range of applications in contrastive learning.

阅读与讨论 → 访问原文 →

10.

arXiv (CS.CL) 2026-06-12 DOI: arXiv:2603.11863

CreativeBench: Benchmarking and Enhancing Machine Creativity via Self-Evolving Challenges

作者:

Zi-Han Wang↗Lam Nguyen↗Zhengyang Zhao↗Mengyue Yang↗Chengwei Qin↗Yujiu Yang↗Linyi Yang↗

The saturation of high-quality pre-training data has shifted research focus toward evolutionary systems capable of continuously generating novel artifacts, leading to the success of AlphaEvolve. However, the progress of such systems is hindered by the lack of rigorous, quantitative evaluation. To tackle this challenge, we introduce CreativeBench, a benchmark for evaluating machine creativity in code generation, grounded in a classical cognitive framework. Comprising two subsets – CreativeBench-Combo and CreativeBench-Explore – the benchmark targets combinatorial and exploratory creativity through an automated pipeline utilizing reverse engineering and self-play. By leveraging executable code, CreativeBench objectively distinguishes creativity from hallucination via a unified metric defined as the product of quality and novelty. Our analysis of state-of-the-art models reveals distinct behaviors: (1) scaling significantly improves combinatorial creativity but yields diminishing returns for exploration; (2) larger models exhibit ``convergence-by-scaling,'' becoming more correct but less divergent; and (3) reasoning capabilities primarily benefit constrained exploration rather than combination. Finally, we propose EvoRePE, a plug-and-play inference-time steering strategy that internalizes evolutionary search patterns to consistently enhance machine creativity.

阅读与讨论 → 访问原文 →

11.

arXiv (CS.LG) 2026-06-11 DOI: arXiv:2606.11711

Capacity-Constrained Online Convex Optimization with Delayed Feedback

作者:

Alexander Ryabchenko↗Idan Attias↗Daniel M. Roy↗

arXiv:2606.11711v1 Announce Type: new Abstract: Online learning with delayed feedback typically assumes that the learner can track all pending rounds until their feedback arrives. In practice, tracking resources are finite, and feedback from untracked rounds is permanently lost. In this paper, we study delayed online convex optimization (OCO) under a hard capacity constraint, where at most $C$ pending rounds can be tracked at any time. To model delay information, we introduce a semi-clairvoyant model that refines the clairvoyant assumption from prior work: rather than requiring delays to be known at prediction time, the learner observes delay expirations online, consistent with the classical unconstrained delayed setting. Our approach proceeds via a reduction to a novel ``delayed and weighted'' OCO problem, using a scheduler that randomizes tracking decisions and importance-weights the resulting observations. For this base problem, we propose and analyze Delayed-Weighted FTRL and its bandit analogue, establishing regret bounds that explicitly characterize the interaction between time-varying weights and delayed feedback. Combining these base learners with our schedulers yields the first regret guarantees for capacity-constrained OCO under convex and strongly convex losses, for both first-order and bandit feedback. For first-order feedback, capacity $C = \Omega(\log T)$ suffices to recover standard delayed OCO rates up to logarithmic factors. For bandit feedback, the regret rates are modulated by powers of $(1 + \sigma_{max}/C)$, where $\sigma_{max}$ is the maximum number of pending observations at any time. This allows the regret bound to degrade gracefully when $C < \sigma_{max}$, while remaining sublinear.

阅读与讨论 → 访问原文 →

12.

arXiv (CS.CL) 2026-06-12 DOI: arXiv:2606.10931

It Takes One to Bias Them All: Breaking Bad with One-Shot GRPO

作者:

Naihao Deng↗Yilun Zhu↗Naichen Shi↗Clayton Scott↗Rada Mihalcea↗

Warning: This paper contains several toxic and offensive statements. Modern large language models (LLMs) are typically aligned through large-scale post-training to ensure fair and reliable behavior. In this work, we investigate how easily such guardrails can be broken by Group Relative Policy Optimization (GRPO). We show that one-shot GRPO training on a single biased example is sufficient to induce systematic bias, with stereotype-driven reasoning generalizing across attributes, categories, and benchmarks. We further find that models differ in their susceptibility based on the initial likelihood of producing biased outputs. Our results reveal a critical vulnerability in post-training: alignment can be overridden by a single example.

阅读与讨论 → 访问原文 →

13.

medRxiv (Medicine) 2026-06-22 DOI: HASH:62663201ba3fce463332c151c3b130a4

Three multimodal large language models fail at clinically actionable breast pathology in three different directions

作者:

Kang↗Y.-J↗Jun↗S.-Y↗Kim↗

Background. Breast cancer treatment depends on histopathological features, such as grade and receptor-defined subtype; however, specialist pathologist access is constrained when the workforce is limited. Commercial multimodal large language models (MLLMs) accept hematoxylin and eosin (H&E) image tiles through paid interfaces without local hardware or fine-tuning. However, prior pathology evaluations addressed only coarse tasks. Whether they reach treatment-determining accuracy and whether vendors agree remain unclear. Methods. We aimed to evaluate three vendor-designated flagship MLLMs (Claude Sonnet 4.6, Gemini 2.5 Pro, GPT-5.5) in 427 invasive breast cancer cases. Each case went to all three with identical H&E tiles and prompts, and the subtype was inferred in the second call. The reference was an institutional sign-out report of an immunohistochemistry-derived subtype. We calculated the concordance, sensitivity, specificity, Cohen's kappa, and pairwise McNemar and Bowker tests. Findings. Claude ranked highest by raw histologic-type concordance but lowest by kappa, classifying all 23 lobular and seven micropapillary carcinomas as invasive breast carcinoma of no special type. The models anchored the Nottingham grade to three modal grades. None of the models reliably identified human epidermal growth factor receptor 2-positive disease. The failure direction was vendor-specific: Claude and GPT-5.5 were under-detected, whereas Gemini was over-called. Twelve prompt variants (4,056 calls) did not recover sensitivity. Interpretation. No current commercial MLLM reaches deployment-ready accuracy for any treatment-determining feature of breast pathology. As each vendor fails in its own fixed direction, changing vendors alters the type of error rather than removing it; therefore, the value of these models is assistive rather than autonomous. At USD 0.20-0.50 per case, they may serve as supervised draft generators that leave the diagnosis with the pathologist.

阅读与讨论 → 访问原文 →

14.

arXiv (CS.CL) 2026-06-17 DOI: arXiv:2606.18021

LegalHalluLens: Typed Hallucination Auditing and Calibrated Multi-Agent Debate for Trustworthy Legal AI

作者:

Lalit Yadav↗Akshaj Gurugubelli↗

AI systems deployed in legal workflows hallucinate at rates that aggregate metrics report at ~52%, but this average conceals where errors concentrate and in which direction they run, leaving compliance officers without an actionable signal for trustworthy deployment. We present LegalHalluLens, an auditing framework with three components: typed hallucination profiles across four legally-motivated claim categories (numeric, temporal, obligation/entitlement, factual) over CUAD (Hendrycks et al., 2021); a Risk Direction Index (RDI) that reduces omission-versus-invention bias to a single deployment-comparable scalar; and a typed debate pipeline calibrated to both magnitudes and directions. Across 510 contracts and 249,252 clause-level instances we measure a within-model gap of approximately 38-40 pp between obligation/numeric and temporal claims that aggregate reporting hides, and show that two systems with matched 52% rates can carry opposite RDIs. The debate pipeline reduces fabricated detections by 45% with per-category gains tracking the diagnosis, matching commercial APIs with a substantially smaller backbone (4B active parameters). Typed profiles and RDI surface failure modes that aggregate metrics hide; we further show these diagnostics serve as calibration inputs for multi-agent debate pipelines, where Skeptic challenges and asymmetric gates targeted at measured failure modes outperform generically-tuned debate. The framework supports direction-aware procurement, accountability, and agent design for legal AI deployed in the wild.

阅读与讨论 → 访问原文 →

15.

arXiv (CS.LG) 2026-06-12 DOI: arXiv:2606.12552

Crossing the Validation Crisis: Cross-Validation Reduces Benchmarking Variance Surprisingly Well

作者:

C\'elestin Eve↗Ga\"el Varoquaux↗Thomas Moreau↗

arXiv:2606.12552v1 Announce Type: new Abstract: Modern machine learning progresses through empirical work, benchmarking new methods to evaluate relative performance. However, the statistical variability inherent to evaluation - exacerbated by the stochastic nature of many algorithms - often makes performance estimation unreliable due to the limited test samples available, leading to a validation crisis in which genuine advances are difficult to discern. In this work, we show that cross-validation improves markedly confidence when evaluating and comparing learning algorithm performances. We introduce the concept of sample gain, which quantifies the virtual data augmentation achieved by using multiple cross-validation splits to reduce benchmarking variance. Experiments on both synthetic and real-world datasets (histopathologic scans and NLP fine-tuning) demonstrate that multiple splits can substantially improve the reliability and stability of performance estimates, with diminishing returns often setting in later than expected. We also introduce a procedure to dynamically early-stop cross-validation by estimating from the first few folds if subsequent folds will bring large sample gains. Our findings highlight the value of pushing cross-validation on available samples to achieve robust and reliable benchmarking.

阅读与讨论 → 访问原文 →

16.

arXiv (CS.AI) 2026-06-19 DOI: arXiv:2606.10358

KG-SoftMAP: Soft Knowledge-Graph Priors for Bayesian Network Structure Learning from Sparse Discrete Data

作者:

Guoliang Xu↗James E. Corter↗

arXiv:2606.10358v2 Announce Type: replace-cross Abstract: Learning Bayesian network (BN) structure from sparse discrete data is hard: when each instance records only a few variables, most variable pairs lack the joint observations needed for reliable scoring, and data-only methods recover little structure. However, imperfect domain knowledge, expressible as a weighted directed knowledge graph (KG), is often available. We propose KG-SoftMAP, which encodes such a KG as a finite-strength, confidence-weighted edge prior and maximizes a MAP objective combining the BDeu score with a logit-form prior; the KG may be expert-curated or LLM-extracted. On synthetic benchmarks with known DAGs, KG-SoftMAP reaches Directed-F1 (DF1) $0.19$–$0.32$ at observation rate $\rho=0.05$ and DF1 $0.44$–$0.97$ at $\rho\geq0.2$, while every data-only learner tested stays near zero under the same sparse masks. Recovery tracks KG quality: controlled corruption degrades it smoothly, a zero-signal KG yields DF1 $0.00$, and a blindly LLM-extracted KG with imperfect precision and recall still drives substantial recovery. On three real sparse educational datasets, the learned BN acts as a concept-level posterior model: on SAF it matches logistic regression (LR) within $0.03$ F1_FAIL while providing an inspectable concept graph, calibrated Fail probabilities, and tractable posterior queries from partial observations.

阅读与讨论 → 访问原文 →

17.

arXiv (CS.AI) 2026-06-11 DOI: arXiv:2603.09555

Compiler-First State Space Duality and Portable $O(1)$ Autoregressive Caching for Inference

作者:

Cosmo Santoni↗Anmol Thapar↗

arXiv:2603.09555v2 Announce Type: replace-cross Abstract: High-throughput Mamba-2 inference is usually tied to fused CUDA and Triton kernels, limiting portability across accelerator backends. We show that the state space duality (SSD) recurrence has a compiler-friendly structure: diagonal per-head dynamics, fixed-size chunking, einsum-dominated compute, and static control flow. Expressing this structure in standard JAX primitives gives a single-source inference path with no custom kernels, a registered JAX PyTree cache, and a compiled on-device autoregressive loop. On a single Google Cloud TPU v6e, batch-1 prefill reaches approximately 140 TFLOPS, or 15% model FLOP utilisation (MFU), the roofline ceiling for this regime, and cached decode reaches up to 64% hardware bandwidth utilisation (HBU). At a 4096-token context, cached decode is 27x–36x faster than full-prefix recomputation across five Mamba-2 checkpoints from 130M to 2.7B parameters. The same source runs unmodified on NVIDIA L40S, where cached decode remains sequence-length independent across all model scales. WikiText-103 validation perplexity matches the Triton reference mamba_ssm v2.2.2 within +/-0.0005 points, and hidden states agree to float32 rounding tolerance. Code is available at https://github.com/CosmoNaught/mamba2-jax.

阅读与讨论 → 访问原文 →

18.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2505.09655

DRA-GRPO: Your GRPO Needs to Know Diverse Reasoning Paths for Mathematical Reasoning

作者:

Xiwen Chen↗Wenhui Zhu↗Peijie Qiu↗Xuanzhao Dong↗Hao Wang↗Haiyu Wu↗Huayu Li↗Aristeidis Sotiras↗Yalin Wang↗Abolfazl Razi↗

Post-training LLMs with Reinforcement Learning, specifically Group Relative Policy Optimization (GRPO), has emerged as a paradigm for enhancing mathematical reasoning. However, standard GRPO relies on scalar correctness rewards that are often non-injective with respect to semantic content: distinct reasoning paths receive identical rewards. This leads to a Diversity-Quality Inconsistency, where the policy collapses into a narrow set of dominant modes while ignoring equally valid but structurally novel strategies. To bridge this gap, we propose Diversity-aware Reward Adjustment (DRA), a theoretically grounded framework that calibrates the reward signal using the semantic density of sampled groups. By leveraging Submodular Mutual Information (SMI), DRA implements an Inverse Propensity Scoring (IPS) mechanism that effectively de-biases the gradient estimation. This creates a repulsive force against redundancy, driving the policy to achieve better coverage of the high-reward landscape. Our method is plug-and-play and integrates seamlessly with GRPO variants. Empirical evaluations on five math benchmarks demonstrate that DRA-GRPO consistently outperforms strong baselines, achieving an average accuracy of 58.2% on DeepSeek-R1-Distill-Qwen-1.5B with only 7,000 training samples and $55 cost, highlighting the critical role of diversity calibration in data-efficient alignment. The code is available at https://github.com/xiwenc1/DRA-GRPO.

阅读与讨论 → 访问原文 →

19.

arXiv (CS.AI) 2026-06-12 DOI: arXiv:2606.13197

ARMOR-MAD: Adaptive Routing for Heterogeneous Multi-Agent Debate in Large Language Model Reasoning

作者:

Fuqiang Niu↗Bowen Zhang↗

arXiv:2606.13197v1 Announce Type: new Abstract: Multi-agent debate (MAD) can improve large language model reasoning, but fixed debate pipelines often waste computation and can amplify correlated errors among similar agents. We propose ARMOR-MAD, a training-free heterogeneous MAD framework that treats debate as conditional computation. ARMOR-MAD combines three components: Pre-debate Agreement Routing (PAR) decides whether independently generated Round-0 answers require debate; Early Agreement Stopping Evaluator (EASE) stops debate after convergence; and Semantic Outlier Detection (SOD) down-weights abnormal final answers during aggregation. Across MATH Level 5, GSM8K, MMLU, and MMLU-Pro, ARMOR-MAD consistently improves over fixed-round heterogeneous debate with the same model pool, reaching 65.5\%, 96.5\%, 90.0\%, and 81.5\% accuracy, respectively. The results suggest that genuine model heterogeneity and agreement-based control are both important for making MAD more accurate and efficient.

阅读与讨论 → 访问原文 →

20.

PLOS Medicine 2026-05-22 DOI: HASH:394c30ca91b2a159bee9b1cba56c0de9

Differences in tuberculosis prevalence by sex in low- and middle-income countries over 1993–2025: A systematic review and meta-analysis

作者:

Nicole A. Swartwood↗

by Nicole A. Swartwood, Nanki Singh, Seyed Alireza Mortazavi, Melike Hazal Can, Hening Cui, Do Kyung Ryuk, Peter MacPherson, Katherine C. Horton, Nicolas A. Menzies Background Global and national initiatives to combat tuberculosis (TB) have expanded over recent years. Despite this, the TB burden remains high in some population groups, with men recognized as having elevated TB risks. Summary measures of sex differences in TB prevalence were last estimated in 2016. Since then, many additional prevalence surveys have been conducted, including in the highest TB burden countries. We conducted a systematic review of sex-stratified TB prevalence survey data published over 1993–2025, to provide updated estimates of male-to-female (M:F) TB prevalence ratios and determine whether sex-related disparities in TB burden have closed over time. Methods and findings We identified surveys reporting community-representative, sex-stratified estimates of pulmonary TB prevalence in low- and middle-income countries (LMICs), including surveys from an earlier review (covering January 1993–March 2016) and a new systematic review (covering 1st December 2015–13th October 2025). This review was prospectively registered with PROSPERO (CRD42024503853) and included searches of PubMed, Embase, Global Health, the Cochrane Library, Africa Index Medicus, LILACS, and SciELO. We extracted data on bacteriologically confirmed and smear-positive TB prevalence among adults (aged ≥ 15 years), stratified by sex. Risk of bias was evaluated using eight criteria specific to prevalence surveys. We fit multi-level Bayesian regression models with study- and country-level random effects to estimate the M:F ratio of TB prevalence (male prevalence divided by female prevalence), overall and for key subgroups. In meta-regression analyses, we estimated how prevalence ratios varied over time and according to known TB risk factors and TB case definitions.We identified 10,124 publications and extracted data from 100 eligible studies representing 102 unique prevalence surveys and 4,658,310 participants (45.6% male) in 33 LMICs. TB prevalence was higher in men than women in 90/102 of the included surveys, with a pooled M:F prevalence ratio of 2.02 (95% credible interval (CrI): 1.71, 2.34) for bacteriologically confirmed TB and 2.38 (95% CrI: 1.91, 2.90) for smear-positive TB. Time trend analyses showed a 2.0% (95% CrI: −0.2, 4.5%) average annual change in the M:F ratio of bacteriologically confirmed TB over the study period. The M:F prevalence ratio was estimated to be higher for countries with greater excess HIV prevalence among men, and countries with greater gender equity (as measured by the United Nation’s Gender Development Index). The estimated M:F prevalence ratio was also higher for surveys that did not restrict testing to individuals reporting TB symptoms. Study limitations include heterogeneity in survey methods and definitions, as well as limited data from the Americas, Eastern Mediterranean, and Europe WHO world regions and post-COVID-19 period. Conclusions Men in LMICs consistently experience TB at a higher prevalence than women. Time trend estimates are uncertain, but consistent with widening sex differences in TB prevalence over the last three decades, despite efforts to address the risk factors underlying this excess TB burden.

阅读与讨论 → 访问原文 →

21.

arXiv (CS.CV) 2026-06-11 DOI: arXiv:2605.30437

Mitigating Content Shift and Hallucination in GenAI Image Editing via Structural Refinement

作者:

Luxi Zhao↗Michael S. Brown↗

Generative AI (GenAI) image editors, such as Nano Banana, produce visually compelling results for retouching tasks, enabling non-experts to edit images through text prompts alone. However, the generative nature of these models often introduces spatial misalignment, texture distortion, and content hallucination, all of which are detrimental to downstream workflows that require pixel-level fidelity. We identify a problem setting we call "structure-preserving GenAI fusion" for black-box GenAI image retouching: retain the perceptual enhancements of a GenAI output while enforcing structural faithfulness to the original input image. To address this problem, we propose a post-processing framework that fuses an input image with its GenAI-enhanced counterpart by first establishing coarse spatial and photometric correspondences, then performing a fusion stage that transfers desired enhancements while suppressing hallucinated content. In the absence of direct prior work in this setting, we evaluate our framework against representative methods from photorealistic style transfer and image fusion. Our experiments demonstrate that our method better preserves aesthetic quality while maintaining pixel-level structural consistency and the input resolution.

阅读与讨论 → 访问原文 →

22.

arXiv (CS.CL) 2026-06-11 DOI: arXiv:2606.11654

The Long Tail, Not the Front Page: Cold-Start Prediction of Crowd Highlight Salience

作者:

Kazuki Nakayashiki↗Keisuke Watanabe↗

A social highlighter's most useful signal – which passages a crowd of readers marks – exists only for documents people have already read. Can the aggregate crowd salience of a document be predicted from its text before its marks accumulate? Prior work on this data found that zero-shot language models recover highlight locations worse than a trivial lead (position) baseline, so we ask whether a model trained on the highlight corpus can beat that baseline. Using a pre-registered ladder of models and a by-document cluster bootstrap, we find a small but robust edge: a logistic ranker over sentence embeddings and positional/contextual features beats the lead baseline by +0.044 average precision (95% CI [+0.029, +0.058]; clears a pre-registered margin delta=0.03 in 97% of resamples, and stable across pipeline re-runs). Two unsupervised extractive baselines (centroid, LexRank-style centrality) lose to lead, and the trained model beats them by +0.108, so the edge is not recovered by generic unsupervised proxies – it reflects learning from real reader marks. In product terms, precision@3 rises from 0.25 to 0.39 (+55% relative) and the model beats lead on 69% of documents. An ablation attributes the edge to the raw embedding (+0.014) and training augmentation (+0.010), each with a positive CI. The edge is not a temporal-generalization failure, and we find no evidence that content drift or near-duplicate leakage explains it. A standardized regression shows the advantage is governed mainly by document popularity (lower popularity, larger edge) and by label reliability. It nearly vanishes only on the most popular content; there it is the lead baseline that strengthens, not the model that weakens. Because our evaluation conditions on documents that eventually accumulated readers, these results are a retrospective cold-start simulation.

阅读与讨论 → 访问原文 →

23.

arXiv (CS.CL) 2026-06-12 DOI: arXiv:2606.13603

Beyond the Commitment Boundary: Probing Epiphenomenal Chain-of-Thought in Large Reasoning Models

作者:

Daniel Scalena↗Sara Candussio↗Luca Bortolussi↗Elisabetta Fersini↗Malvina Nissim↗Gabriele Sarti↗

Chain-of-thought (CoT) reasoning is the dominant paradigm for inference-time scaling in language models, yet the causal influence of individual steps on the final answer poorly understood. We estimate each step's causal importance via early exit and use this measure to study how answers form across the reasoning traces of several model families. Across diverse tasks, we find that reasoning typically crosses a commitment boundary – a sharp transition from transient intermediate guesses to a stable, high-confidence answer. This transition often happens in a single step, well before the model's reasoning block ends, and is followed by epiphenomenal CoT steps that leave the final answer probability unaltered. Using attention probes, we show that answer-formation stages can be linearly decoded from intermediate reasoning steps with high accuracy and generalize robustly to unseen reasoning tasks. We exploit this signal to early-exit reasoning blocks at the commitment boundary, reducing the length of CoTs up to 55\% on average with negligible impact on model performance.

阅读与讨论 → 访问原文 →

24.

arXiv (CS.LG) 2026-06-11 DOI: arXiv:2606.11797

Space-sampled Value Decay: Forgetting Mechanisms for Non-stationary Deep Reinforcement Learning

作者:

Felix St\"orck↗Fabian Hinder↗Barbara Hammer↗

arXiv:2606.11797v1 Announce Type: new Abstract: Studies on rodents such as mice have shown the capabilities to adapt their behavior when dealing with changing parameters (``drift'') of the environment even if no information about change is provided (uncertainty) – a behavior that can be modeled by forgetting mechanisms. Non-stationary Reinforcement Learning (NSRL) deals with adapting state-of-the-art RL methods to deal with changing environments: these however usually require (partially) perfect information about the drift such as ``task IDs'' or ``context''. To mitigate the effects of drift, this work develops Space-sampled Value Decay as an explicit forgetting mechanism for value-based deep RL architectures as a simple yet effective approach. In particular we demonstrate and discuss positive effects but also limitations in achieved returns for modifications of Deep Q-networks (DQN) and Soft Actor-Critic (SAC) when evaluated on non-stationary environments.

阅读与讨论 → 访问原文 →

25.

arXiv (CS.CV) 2026-06-16 DOI: arXiv:2606.16866

Redirecting the Flow: Image Customization through Attention Distribution Shift

作者:

Jie Li↗Suorong Yang↗Jian Zhao↗Furao Shen↗

Subject-driven image customization aims to generate images that not only follow textual instructions but also preserve the identity of a given reference subject. Existing approaches, including test-time fine-tuning, encoder-based methods, and token competition in shared attention spaces, suffer from limited efficiency, misalignment between extracted reference features and the generative process, and interference from irrelevant information. To address these limitations, we formulate the customization task as a distribution shift induced by incorporating reference images into text-to-image generation, and derive a Conditional Attention Distribution Shift formulation grounded in maximum entropy theory. Building on this formulation, we propose CustomShift, a dual-branch architecture based on Stable Diffusion 3. The Reference-Alignment Branch leverages self-attention between reference images and subject names to achieve layer-wise alignment with latent representations, while the Cross-Guidance Branch integrates textual and reference cues to guide generation. Experiments on the DreamBooth and Custom101 benchmarks demonstrate that our method consistently outperforms state-of-the-art approaches, achieving a better balance between semantic fidelity and subject consistency.

阅读与讨论 → 访问原文 →