A study on deep neural network security testing examined four coverage metrics across LeNet, VGG, and ResNet architectures, finding that deeper models and larger datasets do not automatically improve security coverage. T…
Researchers introduced kAgent, an AI-driven tool designed to repair Linux kernel crashes by mimicking human debugging workflows. In tests on the kBenchSyz benchmark, kAgent fixed 54.5% of crashes without localization and…
Researchers introduced MOSAIC, a skill-centric multi-directional planning method that uses physics simulation to improve long-horizon robot manipulation, demonstrating success in both simulation and real-world tests. The…
Researchers developed GenShin, an AI model that predicts protein corona composition on lipid nanoparticles without costly mass spectrometry, potentially accelerating drug delivery innovation. The geometry-enhanced graph …
Prima.cpp, a new software tool, enables large language models with up to 70 billion parameters to run efficiently on ordinary home computers using pipelined-ring parallelism, achieving 5 to 17 times lower token processin…
Researchers have developed a new adversarial attack technique that combines mechanistic interpretability with white-box methods, achieving 80-95% success rates in jailbreaking state-of-the-art large language models like …
Researchers are applying Pareto improvement to counterfactual explanations in AI, making decision-making safer and more transparent by ensuring no trade-offs worsen outcomes. The method, tested on simulated and real data…
Researchers developed MambaCapsule, an AI model that combines Mamba and Capsule networks to diagnose cardiac arrhythmia with 99.54% accuracy on the MIT-BIH dataset and 99.59% on the PTB dataset, while providing transpare…
Researchers introduced the input-skip LBBNN (ISLaB) technique, which reduces neural network complexity by over 99% while maintaining accuracy, achieving 97% accuracy on MNIST with only 935 weights. The method enhances in…
Neural networks and logistic regression models are competing to improve stroke prediction, as strokes occur every 40 seconds in the U.S. and are a leading cause of death. Neural networks offer complex pattern recognition…
Researchers have developed ResBeMF, a new algorithm for AI recommender systems that prioritizes reliability alongside accuracy by providing probability distributions for each recommendation. The model aims to increase us…
Medical Heuristic Learning (MHL), a transparent rule-based approach using large language models, aims to improve clinical decision support by combining high predictive accuracy with interpretability, addressing the opaci…
Researchers found that eight AI models struggle with cultural context, failing to differentiate between cultural groups despite possessing latent knowledge. Disrupting key connections reduced cultural association accurac…
Meta launched an AI tool that lets users alter Instagram content, sparking backlash over concerns it blurs the line between genuine and fabricated posts, threatening user trust and enabling potential misuse for misinform…
Meta Superintelligence Labs launched Muse Spark 1.1, an upgraded coding agent model that scored 69 on the Artificial Analysis Coding Agent Index, approaching GPT-5.5's performance while undercutting competitors on pricin…
The Trump administration approved license-free access for the UAE to buy advanced US AI chips, a decision critics call a 'corrupt deal' after a UAE royal invested $500 million in a Trump family crypto firm. Democratic Se…
Researchers have developed HVR-Met, an AI system that improves extreme weather diagnostics by integrating expert knowledge and a novel feedback loop. The system sets a new benchmark for meteorological AI by focusing on a…
Researchers introduced ARLArena, a framework for stabilizing agentic reinforcement learning (ARL), and SAMPO, an optimization method that addresses ARL's instability. The tools aim to make ARL more reliable for training …
A new Transformer-based model for sign language recognition achieves 88.4% accuracy on unseen classes from the LSA64 dataset using only eight reference examples per class, demonstrating strong one-shot learning capabilit…
A new study demonstrates that discrete diffusion models, when combined with online fine-tuning, replay mechanisms, and validity penalties, outperform traditional methods in molecular optimization across small-molecule an…