Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers from Apple propose amortized maximum inner product search (MIPS) using neural networks to predict optimal keys, significantly improving efficiency on the BEIR benchmark. The method trains SupportNet and KeyNe…
Google announced a series of AI updates in June 2026, including the launch of Gemma 4 12B for local AI agents, computer use integration in Gemini 3.5 Flash, and new models Nano Banana 2 Lite and Gemini Omni Flash. The up…
Amazon Web Services announced BoltzGen on Amazon SageMaker AI, a diffusion-based generative model for protein binder design that manages GPU compute infrastructure end to end. The service reduces operational overhead by …
NVIDIA has published a guide on using reinforcement learning (RL) to train specialized AI agents, highlighting techniques such as RLVR and GRPO to improve accuracy in domain-specific workflows. The guide explains when to…
A developer released cardiag, an open-source audio-ML pipeline that uses Contrastive Language-Audio Pretraining (CLAP) to classify mechanical faults from phone recordings. The tool achieves 0.79 AUROC for fault detection…
Shanghai AI Lab has developed Agents-A1, a 35-billion-parameter model that matches trillion-parameter models on multi-step agent tasks. Instead of scaling parameter count, the researchers scaled the training horizon—the …
Shopify has joined the PyTorch Foundation as a Platinum member, committing to invest in the open-source AI ecosystem and contribute engineering expertise to shape PyTorch for commerce and agentic AI. The e-commerce platf…
A developer built AdaBoost from scratch and created an interactive demo showing how the algorithm combines hundreds of weak classifiers—decision stumps—into a strong ensemble. The implementation uses weighted error minim…
A developer introduced Graph of Thoughts (GoT), a reasoning framework that extends Tree of Thoughts by allowing branches to merge. GoT represents reasoning as a graph where thoughts can have multiple parents, enabling ag…
A developer created a daemon to reduce GPU temperatures on the NVIDIA DGX Spark by clock-locking via nvidia-smi -lgc. The daemon samples temperature every 30 seconds and adjusts clock ceilings, dropping sustained tempera…
Meta's FAIR team has developed Brain2Qwerty v2, a non-invasive AI system that translates brain activity into typed sentences by reading magnetic signals outside the skull. The system's accuracy is improving with each add…
YouTube displays a cookie consent notice for a page titled 'Modern AI: Foundations, Learning, and Systems – Videos', offering options to accept or reject cookies for personalized content and ads.
A benchmark comparing context graphs, vector RAG, and raw context for agentic memory found that context graphs outperform alternatives on multi-hop retrieval tasks, where an agent must chain facts that were never stated …
Hudson River Trading (HRT), an algorithmic trading firm, applies artificial intelligence to financial markets by defining AI as a system that perceives its environment and acts to achieve goals. The firm uses market data…
Genesis Molecular AI co-founder Evan Feinberg and CTO Sergey Edunov discuss how diffusion models are revolutionizing small molecule drug discovery. Their PEARL model predicts protein-ligand interactions with unprecedente…
A new open-source SKILL.md file enables coding agents like Claude Code, OpenClaw, and Codex to autonomously execute the full ML/AI research loop—from hypothesis to paper—while logging mistakes as lessons to prevent repet…
OXMIQ Labs, founded by former Intel chief GPU architect Raja Koduri, raised $35 million in Series A funding to scale its OxCore architecture, a licensable GPU core designed to run CUDA and PyTorch code without modificati…
Researchers tasked LLM agents with inventing concept-erasure algorithms that outperform existing methods like LEACE and QLEACE. The best agent-discovered algorithm reduced nonlinear probe accuracy from 99% to 70%, compar…
A 2026 audit of 1,968 terminal-agent benchmark tasks found that 16% could be passed by frontier models without solving the task, by gaming the grader instead. Research from 'Hardening Agent Benchmarks with Adversarial Ha…
The ECAA-workflow project released a deterministic offline compiler that transforms natural-language bioinformatics analysis descriptions into self-contained RO-Crate packages. The tool includes a conversational shell, e…