Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Google DeepMind released WeatherNext 3 on Thursday, an AI weather model that refreshes forecasts every hour from satellite images, breaking reliance on six-hourly government data cycles and improving temperature, precipi…
A developer's investigation into retrieval leaderboards reveals that rankings are highly sensitive to query wording, meaning models topping benchmarks may not perform best on an application's own data. The analysis shows…
Flipkart's Gridlock Hackathon 2.0, held with Bengaluru Traffic Police, MapmyIndia, Arcadis, and HackerEarth, produced over 1,100 AI and machine learning prototypes to address Bengaluru's traffic congestion, drawing 33,60…
Researchers at Imperial College London's National Heart and Lung Institute have developed Cardiovolt, an AI tool trained on millions of ECGs that can detect hidden signs of heart disease in about two seconds, reporting 9…
A developer has compiled a list of eight open-source tools for building AI agents with memory, including Statewave, Mem0, Graphiti, Cognee, LangGraph, and Letta. The tools are grouped by their function, such as storage, …
Researchers at the University of Tartu used AI to automatically locate field islands across Estonia, identifying nearly 38,000—about 34 percent more than the official register lists. The deep-learning model, trained on o…
Fireworks AI announced the general availability of its Training API and Fireworks Lab, enabling companies to train specialized AI models beyond frontier labs' offerings without building their own training stack. The Trai…
Researchers released VibeVoice-ASR-Streaming, an LLM-based end-to-end streaming speaker-attributed automatic speech recognition system that processes fixed-size audio chunks with lookahead to output 'who said what' in re…
MIT Press has published 'Active Inference: The Free Energy Principle in Mind, Brain, and Behavior', an open-access monograph that presents the free energy principle as a unifying framework for understanding brain functio…
Data Science Weekly's Issue 667 curates links spanning a proposed modern relational query language, a from-scratch walkthrough of ROC-AUC and decision curve analysis, and Observable's relaunch with agent-first computatio…
A user identified as yuil12 on Hugging Face's discussion forum reports that their LoRA (Low-Rank Adaptation) training for AI video models is producing blurry and unrealistic results, and asks how professionals handle LoR…
H Company released NeoMME, a family of 260M and 800M multilingual multimodal encoders that process text and images with a single bidirectional Transformer, trained from scratch with a masked discrete-diffusion objective.…
A developer has built an offline, local edge-compute medical AI system that mines a corpus of 2,500 medical monographs under extreme hardware constraints. The system, designed as a non-commercial proof of concept, integr…
Researchers from Shanghai Artificial Intelligence Laboratory and National University of Singapore introduced Harness-of-Harness (HoH), a framework that enables LLM-based coding agents to autonomously develop software ove…
A new study in Nature Neuroscience used neural networks to show that training on simpler tasks first improves learning in both machines and mouse brains, while poor training causes predictable errors. Researchers at Univ…
A developer successfully ran OpenAI's Whisper large-v3-turbo transcription model and an LLM on an AMD NPU under Linux, achieving a real-time factor of approximately 0.18 on an MSI Stealth A16 AI+ laptop with a Ryzen AI 9…
VLLM's PagedAttention technique reduces GPU memory waste in LLM serving from 60-80% to less than 4% by partitioning the KV cache into non-contiguous blocks, according to a technical analysis. The method enables memory sh…
Andrej Karpathy's Neural Networks: Zero to Hero, the Full Stack LLM Bootcamp by Full Stack Deep Learning, and Stanford's CS336 are among five free courses curated to take learners from LLM beginner to practitioner. The p…
A new machine learning model called the Neural-Curlometer, developed by Delcourt and Mittelholz, uses magnetic field data from NASA's MAVEN satellite to reconstruct large-scale electrical currents in Mars' ionosphere, re…
A developer quantized Qwen3-0.6B's weights to INT8 using per-row scalar quantization, reducing weight matrices from 1761.61 MB to 441.78 MB (3.99x smaller) with worst-tensor relative error of 1.2017%, but found the true …