Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Two startups, San Francisco-based Periodic Labs and Cambridge, Mass.-based Quantum Formatics, are using AI and robots to discover new superconductors that can operate at temperatures above 10 kelvins, the transition temp…
A developer built a Python semantic cache that cuts free-tier LLM token costs by matching new questions by meaning and returning stored answers when embedding distance is close enough. The cache stores past responses on …
A developer's team at an unnamed company built a predictive autoscaling system for GPU workloads on Kubernetes after a production outage caused by reactive scaling. The system uses a Bi-LSTM model embedded in a Kubernete…
A developer survey of AI-assisted monolith-to-microservices platforms in 2026 finds that tooling splits into three roles: boundary planning, code transformation, and analysis. vFunction leads in evidence-based service bo…
StabRise has released @stabrise/scaledp, a document-AI pipeline that runs OCR entirely in the browser via onnxruntime-web, eliminating the need for server-side processing. The new PaddleRecognizer stage decouples text de…
A new e-paper display project by [arnegiacomo] uses a Raspberry Pi 5 and BirdNet-Go, an AI-powered birdsong classifier, to automatically draw pictures of birds heard outside in real time, assembling images from 19th-cent…
An engineer from MonkeyCode warns that timing only the happy path in cache lookups hides the real cost of exception-based miss handling, which can be over 16 times slower and use more memory than a quiet dict.get. The de…
NVIDIA confirmed it will bring DLSS 5 neural rendering to the GeForce RTX 40-Series "Ada Lovelace" graphics cards, after initially focusing on the RTX 50-Series "Blackwell." The company stated that DLSS 5 is its most com…
NanoLM Studio V4, a local desktop workbench for building small decoder-only language models, has been publicly released by its developer. The Tk-based application integrates document ingestion, ByteLevel BPE tokenization…
Dyson unveiled a new lineup of AI-powered appliances at its Dyson Unveiled event in Berlin on Thursday, including the Nurovi robot cleaner family and its first oral-care device, CameraJet, as Chief Engineer Jake Dyson sa…
Post-training, the phase that turns raw AI models into usable assistants, is the overlooked $100 million question in AI development, according to an analysis of the field. The process involves three stages—supervised fin…
A machine learning tutorial explains that traditional machine learning models saturate because they reach theoretical limits and cannot extract more information from additional data, while large neural networks with 20–2…
Engineering teams are shifting from monolithic AI agents to graph-based architectures that define explicit execution graphs with nodes, edges, and shared state. This approach improves debuggability, cost control, and sca…
A Medium article on clinical AI model evaluation argues that F1, AUROC, and calibration answer distinct questions—predict, rank, trust—and that excelling in one does not guarantee the others, warning that a model with ac…
A new arXiv study (2609.02893v1) finds that projecting Llama-3.1-8B-Instruct activations onto a small subset of principal components from the training distribution enables cross-domain deception detection that nearly mat…
A new training-free method called PersonaLink can match retrieval-based personalization on classification tasks but not regression, according to a paper posted on arXiv (2609.02890v1). On 200 users of LaMP-2 (15-way news…
Researchers introduced R²Adapter, a lightweight plug-in Routing and Rewriting Adapter that dynamically allocates queries between vanilla and graph-based Retrieval-Augmented Generation (RAG), reducing graph-based RAG usag…
Researchers introduced BharatGather, a curated dataset of 14,646 records for binary misinformation classification in Indian mass gatherings, built via web scraping of fact-checking platforms, transcript extraction, and L…
Researchers introduced Distilled Rapid Embedding Transfer (DRET), a parameter-efficient method that transfers biomedical knowledge from large models like BioBERT and ClinicalBERT into the 66M-parameter DistilBERT without…
Researchers introduced AdaptiveSpec, a training-free per-step speculative decoding method that adapts both draft verification and tree shape from internal decoding signals, improving throughput over EAGLE-3 by up to 56% …