Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers introduced Tangram, a serving system that enables non-uniform Key-Value cache compression for multi-turn large language model inference. The system uses deterministic budget allocation, head group page cluste…
A developer deploying a 70B Llama-3 model on 8x H100s found that scaling from 8k to 32k context windows causes the KV cache to balloon to 10.7 GB per request, forcing memory paging to CPU at 200 concurrent users. KV cach…
Smart TV owners are unknowingly contributing to the AI data-scraping economy through a software development kit (SDK) from data-collection firm Bright Data, which is embedded in consumer apps and turns devices into resid…
The University of Cambridge and its spin-out DIOSynVax (DVX) Ltd have completed the first-in-human Phase 1 trial of pEVAC-PS, an AI-designed "universal" Sarbeco coronavirus vaccine. The open-label, dose-escalation study …
Researchers have identified exploitable vulnerabilities in the ML-DSA (Module-Lattice-Based Digital Signature Algorithm) implementation, as detailed in a newly released technical paper. The findings highlight security fl…
A developer built an AI-powered medical response system using LangGraph and Python that automates healthcare workflows from heart rate alerts to specialist appointments. The system uses state machines to triage patient s…
Researchers analyzing AI training dynamics found that mixing multiple training objectives under non-stationary distributions produces three distinct behavioral patterns: ecological generalists, conditional policies, and …
Researchers have developed BenchEvolver, a system that automatically generates harder coding benchmarks by first evolving solution algorithms and then deriving new problem statements from them. The method addresses bench…
Researchers have developed Discrete Tilt Matching (DTM), a likelihood-free method for fine-tuning masked diffusion large language models using reinforcement learning. The approach recasts fine-tuning as state-level match…
A final-year mathematics and computer science undergraduate is questioning whether pursuing a career in theoretical robotics, specifically in continual learning for human-like robot adaptation, could inadvertently accele…
A comprehensive guide titled "ML Bible" has been released, covering machine learning topics from classical algorithms to neural networks, transformers, vision, and agents. The resource provides a from-scratch approach to…
After two decades in logistics, supply chain software, and cloud engineering, a developer now writing the Pipeline & Prompts blog argues that understanding problems deeply and translating them into solutions is more valu…
Qualcomm AI Hub provides an end-to-end workflow for deploying machine learning models on Qualcomm devices, as demonstrated in a new tutorial that walks through MobileNet-V2 classification and YOLOv7 object detection. The…
Researchers at CVPR 2026 introduced a training-free single-image diffusion model that generates images matching the patch distribution of a single reference image without neural network training or optimization. The meth…
A developer built a self-improving health platform called Longevity AI that uses five AI agents to automatically update its knowledge base every week without human intervention. The system runs synthetic patient profiles…
A developer has outlined the persistent technical limitations of large language models (LLMs), including hallucination, static knowledge cutoffs, statelessness, limited context windows, and compounding errors in multi-st…
Researchers have developed a Bayesian statistical framework for migrating production systems reliant on large language models when the underlying model reaches end-of-life. The approach calibrates automated evaluation me…
DuckDB has integrated vector and hybrid search capabilities with the Lance lakehouse format, enabling AI workloads directly from SQL queries. SQLite addressed a security vulnerability with a fix for CVE-2025-48595. A new…
NVIDIA released the Jetson Orin Nano Super Developer Kit for $249, offering up to 67 TOPS of AI performance for edge computing and robotics. The tiny dev board runs the full NVIDIA AI software stack and can handle small …
Researchers introduced two new consistency training methods, AttCT and MLPCT, which enforce output consistency in attention weights and MLP post-activations respectively, and applied all four existing and new methods to …