Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Mistral published Shieldstral 1.0 3B, an Apache-2.0 model that classifies text and images against natural-language safety policies instead of a fixed category list. Its July 28 research paper reports 84.9% average F1 on …
Discovery Loop, a new AI research company founded by Jeff Dean, Sanjay Ghemawat, Quoc Le, and Oriol Vinyals, announced its mission to automate scientific and engineering discovery using frontier AI models and large-scale…
Jeff Dean, Google's chief scientist and a key figure in the development of TensorFlow and other AI technologies, announced he is leaving Google to co-found Discovery Loop, a Public Benefit Corporation aimed at automating…
President Donald Trump's 'American Energy Dominance' plan, which supports biomass alongside fossil fuels, is benefiting sustainable aviation fuel (SAF) projects, including the Montana Renewables facility in Great Falls, …
A new dissertation from an unnamed researcher introduces bio-inspired active vision systems that use foveal-peripheral sampling and learned saccades to improve efficiency and robustness in deep neural networks. The syste…
A dissertation from a researcher at an unnamed institution proposes modular parameter-efficient continual learning for Vision Transformers, organizing adaptation into small role-matched corrections rather than full-model…
A new dissertation from an unnamed researcher proposes methods for multi-agent reinforcement learning under information and experience constraints, introducing BAMS and BEPAL for belief-based coordination and EC-HFRL for…
Upstart Holdings Inc. facilitated $4.2 billion in loan originations in the second quarter, a 50% increase year over year, and reiterated its 2026 outlook of about $1.4 billion in revenue and $294 million in adjusted EBIT…
A developer building an entity-matching pipeline across multiple data sources is seeking community input on their dual T4 GPU setup, which uses embeddings for candidate generation and an LLM (Ollama Cloud) for verificati…
Knowledge distillation, a method where a smaller 'student' model is trained to mimic a larger 'teacher' model, can produce a student that matches or beats the teacher on specific tasks, according to an article explaining…
Cisco Secure DDoS Edge Protection introduces a dual-pass machine learning system to counter stealth DDoS tactics, including Pulse Attacks, Carpet Bombing, and Outbound attacks, which evade traditional defenses. The syste…
A Hacker News reader with 50,000+ bookmarks shared a curated list of S-Tier interactive and animated resources for learning computer science, math, algorithms, systems building, networking, graphics, hardware, and AI top…
Sidekick's continual learning loop compresses production experience into model weights, delivering higher quality than frontier models while cutting costs by 96% in Shopify's GraphQL agent. The process starts with defini…
Amazon Web Services announced the general availability of vector search in Amazon DynamoDB, enabling real-time similarity searches on vector embeddings stored alongside operational data with single-digit millisecond late…
Qdrant, an AI vector search database supplier, has applied Google Research's TurboQuant compression technique to embeddings, claiming it achieves competitive recall at 8x compression and beats binary quantization at ever…
A new analysis argues that organ-specific foundation models, trained exclusively on histopathology images from a single organ, may outperform universal models for computational pathology tasks. The author, writing in a t…
Oncotype DX, the most widely used gene expression assay for breast cancer recurrence risk, has been validated in large prospective trials including NSABP-B14, NSABP-B20, TAILORx, and RxPONDER, and its prognostic value su…
Prior Labs' TabPFN, a pre-trained Transformer for tabular data, enables zero-shot machine learning predictions without manual feature engineering or hyperparameter tuning. The model integrates with Scikit-Learn, runs on …
A developer built a Transformer model from scratch to translate English to Sanskrit, training a custom Byte Pair Encoding tokenizer with about 5,000 tokens and a model with roughly 22 million parameters on an RTX 3060 La…
Castform, an AI startup, announced that its reinforcement learning post-training platform combined with Neon's Lakebase Postgres and Search extensions enables open-weights models to match or beat frontier models on agent…