The paper investigates the phenomenon of self‑repair in language models, proposing that it arises from a pre‑existing gain in components that act as counterweights when a component is ablated. By modeling interventions as points on a counterfactual axis, the authors derive an affine law for the causal repair response of fine‑grained units, showing that most downstream directions across several models follow this law. They further demonstrate that the slope of this law can be predicted from fixed weights, suggesting that self‑repair is a predictable, counterweight‑driven response rather than a noisy, unexplained effect.
By Areeb Ahmad, Pratinav Seth, Vinay Kumar Sankarapu
The paper investigates how single‑hidden‑layer MLPs can fit training data yet fail to recover the underlying rule, focusing on higher‑order interactions and nuisance inputs. Using synthetic parity tasks, the authors benchmark different optimizers (SGD, Adam, Muon) and show that while all achieve perfect accuracy on second‑order interactions, performance drops sharply for higher orders, with Muon outperforming the others at fourth order. Experiments also reveal that freezing or removing nuisance‑related weights dramatically alters training outcomes, highlighting the role of nuisance learning in shaping the rules a shallow network can represent.
By Gongyue Zhang, Honghai Liu
The paper evaluates hybrid quantum‑classical machine learning for predicting reduced‑order spatiotemporal brain deformation fields. Using Proper Orthogonal Decomposition to compress high‑dimensional displacement data, the authors compare static temporal‑to‑latent regression and autoregressive latent forecasting models. Classical neural networks outperform all quantum variants, though enhanced quantum circuits improve over minimal ones, indicating that classical architectures still hold a clear advantage in fidelity and stability for this task.
By Tao Liu, Ge He, Dongyu Liang, Wujie Wen
SkillSpec is a two‑phase framework for evolving natural‑language skills in large language model agents. The first phase, consensus‑gated evolution, generates candidate skills from complementary editing intents and commits updates only when paired evaluations reach consensus on overall improvement and non‑negative aggregate gain. The second phase, representation specialization, uses signals from the optimization trajectory to choose an appropriate flat, graph, or hybrid structure for the skill, improving success rates by an average of 6.89% over SkillOpt across six benchmarks and three target language models.
By Huancheng Chen, Xiaodi Sun, Zhaoqiong Huang, Shenyang Huang Shreya Singhal, Jingwen Lu
The paper studies Kolmogorov‑Arnold Networks (KANs), a neural architecture that treats activation functions as learnable components, offering improved interpretability for scientific applications. It investigates how KANs scale with dataset size on image classification tasks (MNIST, Fashion‑MNIST) and a magnetic‑parameter regression task, revealing a broken neural scaling law that transitions from a faster to a slower decay of test loss as data grows. The authors also analyze how the learned activation functions evolve from simple linear approximations to more complex, interpretable symbolic forms as more data is provided.
By Tilen Cadez, Sanghoon Lee, Kyoung-Min Kim
SAGE is a defense against clean‑label data poisoning that relies on a very small set of verified examples—both clean and poisoned—rather than a large clean set. It trains a generic feature extractor on a separate dataset and then uses a non‑parametric, similarity‑weighted prediction to flag poisoned training examples. Experiments on standard benchmarks show that even a handful of verified poisoned examples give a substantial advantage, and that the distribution of verified clean examples across classes is more important than their sheer number.
By Chaeeun Han, Soodeh Atefi, Yevgeniy Vorobeychik, Aron Laszka
ScaffoldM3C is a lightweight, multimodal, auto‑regressive framework that generates stable 3D block constructions by treating the task as a probabilistic next‑block generation problem. It incorporates text, image, and sketch conditioning, introduces a scaffold block token to aid intermediate stability, and uses Sequential Monte Carlo to explore multiple assembly sequences simultaneously. The model is four times smaller than existing baselines, achieving 5‑ to 20‑fold inference speedups while matching or surpassing state‑of‑the‑art construction quality and stability in both simulations and real‑world robot demonstrations.
By Gadiel Sznaier Camps, Chengyang He, Guillaume Sartoretti, Eduardo Montijano, Mac Schwager
SupraTITO is a transferable generative molecular dynamics framework designed for supramolecular systems, specifically peptide self‑assembly. It learns implicit transfer operators conditioned on peptide sequence, molecular topology, and periodic geometry, enabling the propagation of configurations over time intervals far longer than a single MD step. On a dipeptide benchmark, SupraTITO generalizes to unseen sequences, accurately reproduces sequence‑dependent structures and dynamics, and maintains molecular integrity over long rollouts, outperforming direct ensemble prediction models.
By Weilong Chen, Nuno Costa, Julija Zavadlav
Chordonomicon is a new dataset of over 666,000 song-level symbolic chord progressions, each annotated with structural parts such as verse, chorus, and bridge, as well as genre and release date. The dataset was compiled by scraping user-generated progressions from multiple sources and shows strong similarity to established prior datasets. The authors also provide a reproducible benchmark suite for next chord prediction, evaluating RNN, GRU, and LSTM models across various context windows and data scales, and find that structural part annotations consistently improve prediction performance.
By Spyridon Kantarelis, Ioannis Liolitsas, Konstantinos Thomas, Vassilis Lyberatos, Edmund Dervakos, Giorgos Stamou
The paper investigates whether lower prediction error in machine-learning surrogates for fluid dynamics necessarily implies better physical fidelity. Using paired CFD simulations and experimental PIV data for flow around a NACA4418 airfoil, the authors apply a mean‑preserving input intervention that removes velocity fluctuations from selected flow regions. They find that removing fluctuations from the most energetic 10% of cells alters forecasts more than random removal, indicating sensitivity to fluctuation energy. Additionally, a convolutional neural operator (CNO) shows lower overall velocity‑field error but higher two‑component fluctuation‑energy error compared to a reference, and an output attenuation stress test reveals discrepancies between benchmark errors and domain‑summed fluctuation energy. These results suggest that aggregate prediction scores alone are insufficient to assess physical correctness, motivating the inclusion of complementary physical diagnostics.
By Somyajit Chakraborty, Xizhong Chen
RelICL: Training-free Relational Learning with Tabular Foundation Models proposes a new method for relational learning that addresses two key issues of deep feature synthesis—feature explosion and interaction blindness—by propagating and fusing information step by step through the schema graph using a tabular foundation model. The approach retains the benefits of DFS while improving scalability and performance. Experiments on RelBench tasks show that RelICL performs on par with the strongest DFS-based approach.
By Simon Forbat, Rainer Gemulla
SyntheticHLS is a framework that uses large language models to create large-scale, diverse synthetic high‑level synthesis (HLS) datasets. It employs an iterative, feedback‑guided mutation loop that transforms seed designs into more complex, scalable ones, guided by quantitative metrics of design complexity and scalability. The resulting datasets outperform manually curated or zero‑shot generated ones in training deep learning models for HLS quality‑of‑results, offering broader coverage of design space and better generalization.
By Stefan Abi-Karam, Miaoyan Zhou, Callie Hao
The paper presents a finite‑model synthesis that integrates operational state abstraction with optimal control within a stability‑evidence‑revision (SER) framework, termed "Poincaré meets Bellman." It distinguishes qualitative dynamics for reusable action‑response structures from dynamic programming that governs acquisition, retention, reuse, merging, and forgetting, and introduces a Bellman recursion over the joint law of hidden state and deployed memory. The authors derive explicit retention rules, demonstrate how factor sharing and informative observations improve identification, and verify coding and retention calculations through finite enumerations.
By Xin Li
The paper investigates how pre‑training on random input‑output mappings (memorization tasks) can transfer to downstream tasks. It discovers two unexpected patterns: equivalent transfer, where each pre‑training epoch saves roughly one fine‑tuning epoch, and non‑equivalent transfer, where pre‑training on a mismatched task can be more efficient than training directly on the downstream task. Ablation studies reveal that transfer consists of a trivial magnitude‑driven effect in the last layer and a non‑trivial structure‑driven effect linked to covariance in other layers.
By Yimiao Yu, Florentin Guth
The paper introduces Clock Diffusion, a framework for semi‑autoregressive continuous diffusion language models that incorporates position‑dependent noise schedules, efficient training, and sampling algorithms. It presents two generation modes—block and sliding window—and defines ClockDLMs, a family of Gaussian models that achieve state‑of‑the‑art diffusion likelihoods on OpenWebText and outperform continuous baselines on GSM8K while matching or exceeding discrete diffusion models. The authors also propose Cache Grab, a set of efficient samplers that adapt accelerated inference techniques from discrete diffusion to further improve model quality and efficiency.
By Yair Schiff, Omer Belhasin, Roy Uziel, Matan Rusanovsky, Ran Zilberstein, Marianne Arriola, Gilad Turok, Guanghan Wang, Volodymyr Kuleshov, Michael Elad
The paper introduces Evidence‑Gated Research (EGR), a statistical adoption layer designed for adaptive model search where each new challenger is frozen until evidence is revealed. EGR constructs anytime‑valid evidence across predefined environments, routes it toward unresolved components, composes a persistent candidate e‑value, and forwards this to an online controller. The method guarantees false discovery rate control for the declared all‑environment adoption target, even when earlier adoptions influence later challengers, and demonstrates evidence efficiency and persistent FDR control in large‑scale benchmarks.
By Yifan Guo
The paper introduces higher-order positional encodings that enrich graph representations by incorporating topological information from lifted incidence structures, without altering existing graph learning backbones. It theoretically shows that these encodings can mix graph Laplacian frequencies beyond what scalar spectral filters achieve, and demonstrates their effectiveness on Graph Transformers for datasets like ZINC and synthetic benchmarks. The approach bridges graph positional encodings and topological deep learning, enabling standard models to exploit higher-order interactions.
By Caleb Stam, Aagrim Hoysal, Sanjukta Krishnagopal
The paper demonstrates that reinforcement learning with verifiable rewards (RLVR) can improve vision‑language benchmark performance even when models are trained without visual input. When real images are introduced at test time, models trained blind recover about half of the performance gain at 3B parameters and nearly four‑fifths at 7B, but extended real‑image training can erode grounding while benchmark gains persist. The authors propose a visual resolvability rule and show that requiring visual evidence for correct answers leads to significant improvements in target discovery and generalization to unseen question types, while controls confirm that the gains stem from actual visual grounding rather than artifacts.
By Haocun Ye, Xinlong Jiang, Qile Chen, Bingyu Wang, Teng Zhang, Shubai Chen, Tingyu Wu, Zhenkun Zheng, Yiqiang Chen
The paper reports a large-scale, compute-controlled study of Chemical Language Models (CLMs) involving over 30,000 experiments across different molecular representations, tokenizations, model sizes, datasets, and architectures. It finds clear scaling trends in pretraining loss but shows that these improvements do not translate into proportional gains in goal-directed molecular design, with chemical syntax saturating early while semantic properties develop more slowly. The authors release a new suite of models, NovoMolGen, that achieves state-of-the-art results in drug discovery tasks, highlighting a disconnect between representation learning and downstream design and calling for new pretraining paradigms that target chemical semantics.
By Roshan Balaji, Kamran Chitsaz, Quentin Fournier, Nirav Pravinbhai Bhatt, Sarath Chandar
The paper evaluates System One decision models—typed models that output probabilities for branching decisions—against supervised classifiers and generative language models on automated decision gate tasks. Eight checkpoints from six families, including the hosted model Jev, were benchmarked on workflow, intent, and social‑science items, showing that small trained classifiers match or slightly outperform decision models on intent and workflow when labels are available, while decision models outperform zero‑shot classifiers when labels are absent. The study also explores calibration, risk thresholds, and cost‑efficiency trade‑offs, providing condition‑dependent design guidelines for automated decision gates.
By Amir Rafe, Subasish Das