The paper introduces PureVision, a geometry‑supervised framework for medical vision‑language models that tackles the challenge of interpreting multi‑phenotype lesions. It comprises PureEyes, which imposes geometric constraints to preserve anatomical hierarchies and phenotypic subclass relationships, and PureNeurons, which aggregates lesion‑specific anatomical and phenotypic evidence in a learned latent space. Experiments on LIDC‑IDRI, CBIS‑DDSM, and 3DReasonKnee show that PureVision enhances lesion grounding and phenotype characterization in visual question answering and radiology report generation.
By Hao Wang, Qiwei Zeng, Shuchang Ye, Jinghao Lin, Yuezhe Yang, Yige Peng, Jinman Kim, Lei Bi
The paper introduces CHASE, a framework that leverages Geometric and Spectral Alignment (GSA) to inform practical model design. GSA characterizes trained networks via spectral concentration, channel alignment, support structure, and singular basis changes, and CHASE applies these insights to six applications including parameter‑efficient fine‑tuning, structured pruning, and model merging. New methods—CAGA, SAKV, and CAPS—use GSA to share attention heads, compress KV caches, and prune structured outputs, demonstrating improved performance over baselines.
By Wei Wang, Wei Jiang, Ziran Liu
The paper proposes a framework that adapts systematic review methods—planning, conducting, and documenting—to the retrieval of machine learning assets in AI registries. By treating models, datasets, and benchmarks as first‑class units of analysis, the framework incorporates registry‑aware search, cross‑registry schema alignment, and dependency‑driven exploration to make asset selection transparent, reproducible, and evidence‑based. The authors argue that this approach transforms ML asset discovery from an informal, ad‑hoc process into a structured, systematic practice.
By Alexandra Gonz\'alez, Quim Motger, Xavier Franch, Silverio Mart\'inez-Fern\'andez
The paper introduces COPC, a Coupled Off-Policy Correction method for asynchronous reinforcement learning of large language models. COPC coordinates policy-side and advantage-side corrections by combining token-level ratio masking with two-sided clipped-ratio weighting of TD residuals, addressing both policy mismatch and advantage staleness. Experiments show COPC outperforms existing asynchronous baselines on tool-integrated mathematical reasoning and search tasks, while maintaining training stability and minimal overhead.
By Zicheng Hu, Zhijian Zhou, Xuan Zhang, Yuchen Liu, Cheng Chen, Yuan Li, Qi Gu, Yan Feng, Hongyan Hao, Chao Qu
The paper investigates why large neural networks outperform smaller ones, focusing on a phenomenon called primacy bias—where early data distributions hinder learning of later ones. It shows that overparameterized models are more robust to this bias, especially during sequential pretraining on heterogeneous data. The authors propose Exposure Therapy, a regularization technique that improves small models’ performance on late data and overall capabilities up to the billion‑parameter scale.
By Mohnish Harwani, Yujia Zheng
arXiv:2610. 09675v1 Announce Type: new Abstract: We study the accuracy of Gauss-Newton curvature in ridge-regularized nonlinear least squares.
By Kihun Rhee, Hanjoon Byun
MUNITE is a latent‑variable framework that unifies encoding and latent generation for any‑to‑any multimodal generation. By treating different amounts of observed evidence as varying inference problems, it models the conditional distribution over a shared latent representation, using a single conditional flow model. The approach employs self‑distillation with conditional flow matching to learn from incomplete data, achieving competitive or superior generation quality and coherence across datasets such as PolyMNIST‑D‑Q, FFHQ64, and image‑text‑audio.
By Kyeongmin Yeo, Minhyuk Sung
The paper introduces Interaction-Constrained Drive Policy (ICDP), an offline reinforcement learning framework for autonomous driving that explicitly controls interaction-level distribution shift. ICDP decomposes joint-support degradation into ego-support and interaction-support components, estimating the latter via contrastive density-ratio estimation without requiring joint-density modeling or simulator rollouts. Experiments on nuPlan, Interplan, and real-world truck data demonstrate that ICDP reduces selection of high-value but interaction-unsupported trajectories and improves performance in interaction-critical driving scenarios.
By Mahmoud Selim, Cristina Cipriani, Karl Henrik Johansson
arXiv:2610. 09838v1 Announce Type: new Abstract: We present a framework for building and interpreting minimal transformer models.
By Raneem Mahajne, Toviah Moldwin
The paper presents a nonlinear dissipative dynamical model of human learning, featuring a concept‑transfer matrix, per‑concept forgetting rates, and a saturating practice‑response gain. It proves a structural identifiability theorem that guarantees exact parameter recovery under explicit excitation conditions, provides closed‑form recovery for the two‑concept case, and introduces a fast, L‑stable numerical scheme enabling estimation on large learner cohorts. Empirical results show exact recovery on synthetic data with designed excitation, but failure to recover on large observational benchmarks, illustrating a sharp boundary between identifiable and unidentifiable regimes and offering a validation protocol for interpretability claims.
By Arman Kostanian, Armen Beklaryan
The study investigates whether a biological wiring diagram can serve as a computational substrate for tasks beyond its evolutionary purpose. Using the MaleCNS v1.0 Drosophila connectome as a fixed recurrent network, the authors train models to perform bounded addition and a grounded relational language task, achieving high accuracy (92.77% for addition and 61.59% for language) that surpasses degree-preserving rewired controls. The results demonstrate that the higher-order structure of the MaleCNS provides a reusable inductive bias for these cognitive tasks.
By Joonghui Cho, Minchan Kang, Daeshik Kim
The paper presents upper and lower bounds on the regret for m‑set adversarial bandits across various utilities—winner reward or sum of rewards—and feedback models, including winner index, winner reward, sum of rewards, and their combinations. By comparing these results to standard bounds for combinatorial and MNL bandits, the authors demonstrate that small changes in the setting can dramatically affect learning rates. The main technical contribution is the derivation of information‑theoretic lower bounds, and synthetic experiments validate the theoretical findings.
By Nicol\`o Cesa-Bianchi, Matteo Papini
The paper introduces a new class of offline adversarial attacks on warm‑start bandit algorithms with bounded rewards. Unlike prior methods that suppress non‑target arms, the authors show that promoting the target arm is essential when it lies near the lower reward boundary, and they design an attack that achieves optimal sublinear cost by balancing target promotion and non‑target suppression. The attack is extended to several popular bandit algorithms and validated experimentally on real‑world and synthetic data.
By Qirun Zeng, Manhin Poon, Xiangxiang Dai, Qixin Zhang, Jinhang Zuo
arXiv:2610. 10246v1 Announce Type: new Abstract: Constant-step stochastic approximation generally has a nonzero stationary mean error that persists under time averaging.
By A. Ch. Madhusudanarao, Rahul Singh
Efficient Provably Private Classification with a Tabular Foundation Model introduces PrivTab, a tabular foundation model that embeds differential privacy directly into its architecture. Pretrained on simulated datasets, PrivTab uses in-context learning to convert sensitive rows into compact, provably private summaries, enabling a single forward pass for classification. The model outperforms private linear and neural‑network baselines under moderate-to-strong privacy, shows negligible membership leakage, maintains well‑calibrated predictions, and reduces dataset fitting time by 10,000×.
By Talal Alrawajfeh, Cristiana Diaconu, Ossi R\"ais\"a, Sebastian Rodriguez Beltran, Yuan He, John Bronskill, Richard E. Turner, Antti Honkela
The paper introduces the Adversarial Surface-Form Robustness Dataset (ASRD), which contains 2,100 prompts across seven surface‑form families to test large language models on non‑canonical inputs. Five open‑weight models were evaluated on 10,500 responses, and a Quad‑State Evaluation Rubric categorized each response as harmful compliance, safe response, comprehension failure, or indeterminate. Results show that emoji and invisible Unicode variations produce few comprehension failures, while leetspeak, encoded wrappers, and hybrid transformations lead to high comprehension failure rates and varied harmful compliance percentages.
By Pavan Maddula
ORDERS is a personalized federated learning configuration that combines a shared backbone, private residual adapters, and a classifier, using geometric weights based on descending update norms, feature alignment, and private‑parameter perturbations. In extensive experiments on CIFAR‑10 and Sent140, ORDERS achieves slightly higher mean client accuracy than baseline methods, with a 5.47% reduction in parameter payload. Ablation studies show limited, dataset‑dependent benefits from norm ranking and alignment, and no clear advantage from perturbations.
By Koffka Khan
The paper introduces a scalable heterogeneous Graph Neural Network framework that automatically generates and evaluates shared roadmaps for multi‑agent path planning in continuous environments. By representing waypoints, agent locations, and task locations as distinct nodes, the GNN learns from expert solver trajectories to identify critical points and prune redundant connections, producing a compact, coordination‑aware roadmap. Experiments show the approach can reduce planning effort, achieving at least a 40% reduction in runtime and graph size for dense roadmaps.
By Brandon Ho, Nikola Rogers, Seung-Kyum Choi
Kernel Autoresearch (Kernaut) is a new approach that treats kernel design as open‑ended model discovery, using coding agents to write kernel programs while construction contracts guarantee validity. The method maintains a quality‑diversity archive and novelty screening to encourage functionally distinct kernels, which are shown to generalize to unseen tasks and outperform meta‑learned deep kernels on black‑box optimization families. Additionally, kernels discovered from enzyme‑kinetic rate laws achieve lower error than tuned ARD and deep kernel baselines, and a human‑refined version further improves predictive accuracy and optimization regret.
By Richard Cornelius Suwandi, Feng Yin, Kevin Murphy
The paper addresses the bias introduced by two‑level softmax (2LS) sampling, which partitions items into clusters and samples a cluster before an item. It identifies that 2LS misweights clusters by ignoring size imbalance and intra‑cluster similarity dispersion, leading to systematic sampling bias. The authors propose two corrected methods, Size‑Corrected 2LS (S‑2LS) and Size‑ and Dispersion‑Corrected 2LS (SD‑2LS), that eliminate these biases with negligible computational overhead and demonstrate improved sampling performance on five large‑scale datasets.
By Walid Bendada, Guillaume Salha-Galvan