Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
A team of engineers at the Micro Air Vehicle Lab at Delft University of Technology won a high-profile Abu Dhabi first-person-view (FPV) drone race in November 2025 with an AI-piloted quadcopter, the first time an AI-cont…
FineEnvs founder Adithya S Kolavi released two reinforcement-learning environments for OCR and speech recognition on October 14th, containing more than 12 million tasks across 22 OCR languages and 102 speech languages, a…
Researchers primarily from Nobel laureate David Baker's lab at the University of Washington used AI-based protein design tools to build new proteases by taking the catalytic core of two protease types — zinc proteases an…
A developer resolved the TensorFlow TypeError "Using a `tf.Tensor` as a Python `bool` is not allowed" in the grad_cam exercise of the AI for Medicine specialization's Course 3, Week 3, Exercise 1 by replacing Python cond…
The Allen Institute for AI (Ai2) released HiRO-ACEv2, a climate emulator that generates global 3 km-resolution rain, wind, and pressure data at roughly 1/300th the cost of the X-SHiELD high-resolution atmospheric model. …
An AI postprocessing system developed by MIT climatologist Judah Cohen and colleague Lester Mackey doubled overall prediction accuracy for the two-to-six-week subseasonal forecasting window, the team reported in a study …
A single end-to-end reinforcement learning policy completed all 20 assembly tasks on the NIST board in simulation with 98.2% success, according to AlphaNIST, which trained the policy from binary task rewards without pres…
Physical Intelligence launched a beta of its Pi API, letting developers upload their own data, run fine-tuning on Physical Intelligence-managed compute, and access fine-tuned versions of its π0 and π0.5 robotic foundatio…
A 13 October 2026 arXiv paper introduces Tev-Gym, described as the first long-horizon agent training dataset with verifiable, dense, progressive rewards, comprising 1,170 verified task environments across 7 domains. The …
A team led by Wenyi Li submitted EmbodiedEvo, a self-improving coding agent for physical tasks, to arXiv on 13 Oct 2026 (latest version 14 Oct 2026), reporting an average score of 44.48 and a 37.82% success rate on RoboD…
Scikit-LLM, a Python library that wraps pre-trained large language models behind the scikit-learn fit() and predict() interface, enables few-shot text classification with as few as 3 to 5 labeled examples, according to a…
Topos Bio, an AI-powered biotechnology company, introduced two AI foundation models — Topos-2 and Topos-Bind — to predict how intrinsically disordered proteins (IDPs) move and change shape and how those dynamic proteins …
Amazon SageMaker JumpStart added two capabilities in 2026: pre-configured endpoint deployment options optimized for low cost, low latency, or general-purpose workloads across generative AI models including Neuron-optimiz…
Developer Boobalan Arjunan released WaveWash, an open-source Python package in early alpha (v0.1.0a1) that cleans, repairs, and dedupes audio datasets before speech-model training for Whisper, TTS, and voice cloning. Wav…
GitHub trained a credential classifier for its secret scanning pipeline that cut median post-push scan execution time by 86% and model serving costs by more than 98% compared with its previous large language model pipeli…
Training GPT-2 only on its own output for four rounds made it 52% worse at predicting unseen Wikipedia text, while adding 50% new real data each round limited the degradation to 6%, according to experiments cited in a gu…
Liquid AI released LFM2.5-24B-A2B, an open-weight 24B Mixture-of-Experts reasoning model with 2B active parameters aimed at on-device agents. The model scores 48.09 on LiveClaw and decodes 184 tokens/s on an Apple M5 Max…
Researchers at the Indian Institute of Science (IISc) used machine learning combined with quantum mechanics simulations to expand a copper-catalyst CO2 hydrogenation reaction network from 152 reactions to 9,389, accordin…
SK hynix researchers Won-Tae Koo and Jihun Kim presented a co-optimization approach for cell, non-cell, and page schemes in 3D FeNAND-based analog compute-in-memory (A-CiM) at the 2025 IEEE International Electron Devices…
Liquid AI's LFM2.5-24B-A2B, a 24-billion-parameter mixture-of-experts model that activates only about 2.45 billion parameters per token, scored 48.09 on the LiveClaw agentic and tool-use benchmark versus 38.19 for Qwen3.…