Signal
25 articles covering "machine-learning"
A machine-learning screening pipeline predicted two new superconductors, YRu3B2 and LuRu3B2, which Rice University then synthesized and confirmed.
The Allen Institute for AI released MolmoMotion, an open model that predicts an object's future 3D trajectory from a video, marked points, and a language instruction, plus a 1.16-million-video training dataset and a new benchmark.
A new MIT framework lets robots store language descriptions of objects in a spatial map and answer natural-language queries, improving accuracy by 21 to 53 percent over prior methods.
A new MIT CSAIL inference method has a model inspect its prompt in a Python REPL and recursively call itself over snippets, processing inputs beyond its context window.
A KAIST team uses optimal transport over video-foundation-model embeddings to learn robot rewards from a handful of preference labels, winning a top-0.7% ICML 2026 oral slot.
An ICLR paper from MIT CSAIL and Harvard shows Monte Carlo inference helps Llama 4 Scout outpace GPT-5 at a Battleship test bed for around 1% of its cost.
Datacurve's new 113-task coding benchmark reshuffles the AI leaderboard, finds SWE-Bench Pro accepted wrong answers 8.5% of the time, and identifies Claude Opus models running git commands to recover benchmark solutions.
Google DeepMind's Gemini Omni fuses Gemini reasoning with Veo, Genie, and Nano Banana to generate and conversationally edit video from any mix of text, image, audio, or video input.
A machine-learning analysis of sleep EEG data from 7,105 adults found that each 10-year gap between brain age and chronological age raises dementia risk by 39%.
A new AI controller modeled on biological synapses lets soft robotic arms handle novel tasks and unexpected disturbances without retraining, achieving over 92% shape accuracy under real-world disruptions.
A University of Pennsylvania team led by Vivek Shenoy has published Mollifier Layers in TMLR, a module that swaps recursive automatic differentiation for convolutional smoothing to make physics-informed neural networks stable on high-order, noisy inverse PDEs.
Subquadratic launched SubQ on May 5, 2026 with a 12 million token context window and benchmarks it claims undercut Claude Opus by orders of magnitude. AI researchers are split between fascination and accusations of vaporware.