AI/ML arXiv cs.AI

Harnessing X-ray Absorption Spectroscopy Data through Multimodal Mining of Battery Literature

A multimodal mining pipeline that converts X-ray absorption spectroscopy data from battery literature into a structured, AI-ready open dataset.

AI/ML arXiv cs.AI

Towards simultaneous decoding of kinetic and kinematic movement parameters during grasp and lift task by noninvasive brain imaging

Comparison of regression models for brain-machine interfaces, showing that attention-based regressors excel at simultaneous multi-parameter movement decoding from EEG signals.

AI/ML arXiv cs.AI

Exact Symmetry as Algebra: A Machine-Verified Tensor Calculus that Enforces Physical Selection Rules

Researchers introduce a machine-verified tensor calculus in Lean 4 that enforces exact physical symmetry in ML models, eliminating equivariance errors.

AI/ML arXiv cs.AI

VistaHop: Benchmarking Long-Horizon Visual DeepSearch

VistaHop and VistaArena are introduced as benchmarks to evaluate multimodal LLMs' ability to perform long-horizon visual search and iterative reasoning.

AI/ML arXiv cs.AI

An Empirical Audit of Input Encoders for Multi-Channel Signal Transformers

An empirical study of input encoders for multi-channel signal Transformers suggests that standard per-channel linear projections are practically equivalent to more complex alternatives.

AI/ML arXiv cs.AI

Minimax-Optimal Generalization Bounds for Smooth Deep Neural Networks Trained by (Stochastic) Gradient Descent

This paper provides minimax-optimal generalization bounds for smooth deep neural networks trained via GD and SGD, linking them to kernel methods.

AI/ML arXiv cs.AI

SafeECGMatch: Calibration-Aware Joint Frequency and Time Space Semi-Supervised Learning for Open-Set ECG Classification

SafeECGMatch is a semi-supervised learning framework for ECG classification that uses calibration-aware techniques to handle out-of-distribution anomalies.

AI/ML arXiv cs.AI

Ouroboros-Spatial: Closing the Data-Model Loop for Spatial Reasoning

Ouroboros-Spatial is a self-evolving training framework for MLLMs that closes the data-model loop to improve spatial reasoning efficiency.

AI/ML arXiv cs.AI

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

FCGraft reduces latency and improves robustness in embodied agent policies by grafting validated function-level KV caches instead of full re-computation.

AI/ML arXiv cs.AI

Keyless Attention: Value-Space Routing and Value-Only Caching for Efficient Transformers

Keyless Attention proposes a value-space routing mechanism that eliminates key representations, reducing KV-cache memory and access overhead by 50%.

AI/ML arXiv cs.AI

Revealing Hidden Model Behaviors with Task-Specific Self-Reports

SAR is a lightweight LoRA adapter designed to audit fine-tuned models by making them report their own hidden behaviors in plain language.

Cybersecurity arXiv cs.AI

G2VD: Generalizable AI-Generated Video Detection via Counterfactual Intervention and Causal Disentanglement

G2VD is a generalizable AI-generated video detection framework using counterfactual intervention and causal disentanglement to avoid shortcut learning.

Other Hacker News

The First Transatlantic Telegraph Cable Was a Bold, Beautiful Failure

A historical account of the first transatlantic telegraph cable and its initial failure.

AI/ML arXiv cs.AI

Adaptively Robust LLM Monitoring via Activation Watermarking

Proposed Activation Watermarking (AWM) to prevent adaptive attackers from evading LLM monitoring via hidden state randomization.

AI/ML arXiv cs.AI

GroupRAG: Cognitively Inspired Group-Aware Retrieval and Reasoning via Knowledge-Driven Problem Structuring

Introduces GroupRAG, a framework that uses knowledge-driven keypoint grouping to improve LLM retrieval and reasoning based on cognitive science.

AI/ML arXiv cs.AI

REAP: Automatic Curation of Coding Agent Benchmarks from Interactive Production Usage

Presents REAP, an automated pipeline for curating production-derived benchmarks for AI coding agents, resulting in the Harvest benchmark.

AI/ML arXiv cs.AI

Shot-based quantum encoding: a data-loading paradigm for quantum neural networks

Introduces shot-based quantum encoding (SBQE) to efficiently load classical data into quantum neural networks without data-encoding gates.

AI/ML arXiv cs.AI

The Fast Lane Hypothesis: Von Economo Neurons Implement a Biological Speed-Accuracy Tradeoff

Develops the Fast Lane Hypothesis and a computational model for von Economo neurons to explain biological speed-accuracy tradeoffs in social cognition.

AI/ML arXiv cs.AI

Facial-Expression-Aware Prompting for Empathetic LLM Tutoring

Explores the use of facial-expression-aware prompting to increase empathy and responsiveness in LLM-based tutoring systems.

AI/ML arXiv cs.AI

BioHiCL: Hierarchical Multi-Label Contrastive Learning for Biomedical Retrieval with MeSH Labels

Introduces BioHiCL, a hierarchical multi-label contrastive learning approach for improving biomedical information retrieval using MeSH labels.