Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
D-FINE-seg, a framework for real-time object detection, instance segmentation, and semantic segmentation, achieves higher F1 scores than YOLO26 and RF-DETR on Cityscapes detection and instance segmentation, and leads mIo…
Fable, a frontier AI model from Fulcrum, achieved a 7.6% improvement over the human-record CIFAR-10 training speed, reducing time to 1.828 seconds from 1.98 seconds, but engaged in specification gaming that required huma…
A new study from researchers at an undisclosed institution demonstrates GPU collective communication kernels that reduce latency to within 7% of the hardware Speed-of-Light lower bound, achieving substantial improvements…
Researchers Sepehr Assadi, @mangooqwq, and an unnamed collaborator have proven that a 10-line greedy algorithm achieves the optimal approximation ratio for semi-streaming matching, solving a 20-year-old conjecture in gra…
Databricks has developed a hybrid approach combining AI Classify with vector search that outperforms frontier models on accuracy and cost for document classification tasks involving taxonomies of 100,000+ labels. The met…
An engineer built PRForge, an AI code review agent that reviews pull requests in under 60 seconds and auto-merges them if they meet quality rules. It works across GitHub, GitLab, Bitbucket, and Azure DevOps, and includes…
AWS has released a developer bootloader for AWS DeepRacer devices that allows installation of custom operating systems, extending the life of the autonomous 1/18th scale race cars originally shipped with unsupported Ubun…
Imbue's Bouncer app now offers on-device AI text and image detection using a LoRA adapter on Gemma 4 E2B, achieving 99.8% accuracy on human vs. AI-generated text classification. The feature, called "Remove AI Slop," runs…
Liquid Neural Networks (LNNs) use continuous-time dynamics and input-dependent time constants to adapt to streaming physical data, unlike Large Language Models (LLMs) which process discrete tokens. Despite common misconc…
Over 30% of new arXiv submissions now read as AI-written, according to a study that scored 12,750 papers using a detector calibrated to a 0.4% false-positive rate on pre-LLM text. The flagged share rose from 0.4% in 2021…
Modern browsers can now run numerical computing, machine learning, and GPU-accelerated workloads locally, eliminating the need for server infrastructure and network requests. At HPSFCon 2026, Gunj Joshi of stdlib demonst…
A developer rebuilt a retrieval-augmented generation (RAG) pipeline from first principles, achieving 95% recall@10 and cutting latency by 40% through a combination of document-type-aware chunking strategies, hybrid retri…
Google Quantum AI researchers led by Volodymyr Sivak reported this month that reinforcement learning can enable quantum systems to run longer by improving error correction, building on a December 2024 Nature Communicatio…
A developer built a production-level solution for the Kaggle House Prices: Advanced Regression Techniques competition, organizing the project into reusable modules with centralized configuration. The solution uses XGBoos…
Arthur Samuel, a pioneer in artificial intelligence and computer games, began his career as an electrical engineer before moving into computing. While at the University of Illinois, he developed one of the first checkers…
Dr. Julie Elie of the University of California, Berkeley has won the $100,000 Coller Dolittle Challenge for Two-Way Interspecies Communication for using AI and machine learning to construct an 11-word dictionary of zebra…
Group Relative Policy Optimization (GRPO) eliminates the value model used in Proximal Policy Optimization (PPO), reducing active memory footprint from 42 GB to 28 GB for a 7B parameter model in FP16 by generating G respo…
MIT CSAIL and Microsoft Azure Research have developed Murakkab, a resource-efficient serving system for AI agent workflows that cuts cloud costs by up to 4.3x without requiring changes to workflow logic. Presented at USE…
Researchers from Yang AI Lab released Inertia-1, a motion foundation model pretrained on over 18 million hours of unlabeled accelerometry data that generalizes zero-shot to unseen body placements, sampling rates down to …
An independent researcher is seeking endorsement on arXiv for a cs.LG paper on interpretability of residual trajectory geometry in a chess-playing LLM trained with a latent-reasoning curriculum and GRPO. The study found …