Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
DeepMind released DiffusionGemma, a new large language model that uses discrete diffusion to generate entire sequences in parallel instead of autoregressive token-by-token generation. The model achieves over 1,000 tokens…
Mixture-of-Experts (MoE) architecture allows large language models to use only a fraction of their total parameters for each token, enabling models with 671 billion total parameters to run at speeds comparable to a 37-bi…
Researchers have developed a new global dataset of annual migration flows across 230 countries from 1990 to the present, using deep learning to integrate diverse data sources. The model, which outperforms existing five-y…
NVIDIA's cuDF library now provides GPU-accelerated execution engines for the Polars Lazy API, enabling users to run dataframe operations on GPUs with automatic fallback to CPU when operations are unsupported. The engine,…
A developer achieved 775 tokens per second running the full BF16 DiffusionGemma model on an Nvidia RTX 6000 Pro using a Red Hat fork of vLLM, demonstrating extremely fast local AI inference at short contexts. However, pe…
A developer built DetectNix Vision, a Windows desktop application that performs local AI-powered image analysis without uploading user data to the cloud. The project required solving significant engineering challenges in…
GKE Labs has launched OpenRL, an open-source, self-hosted training API for fine-tuning large language models on Kubernetes. The API decouples post-training infrastructure from reinforcement learning research workflows, u…
Amazon published a how-to demonstrating an intelligent document processing pipeline on Amazon Bedrock that combines on-demand FIFO queue processing for low-latency single documents with batch inference for asynchronous, …
L&T Technology Services (LTTS) announced a strategic go-to-market partnership with Databricks to co-develop Industrial AI solutions for asset-intensive industries. The collaboration will combine LTTS' domain expertise ac…
An arXiv preprint submitted 23 Oct 2025 evaluates a topic-aware, hierarchical summarization pipeline that applied Latent Dirichlet Allocation to 50 transcribed stories from African American storytellers, identifying 26 t…
ServiceNow Chief Learning Officer Jayney Howson rebuilt the company's training program to be AI-native, using Synthesia to produce over 5,000 videos in 18 months and reducing course production time by roughly ten times. …
Medical device recalls hit a four-year high in 2024 with 1,059 events recorded in the US, and roughly 25% trace back to mislabeling of batch numbers and expiry dates on packaging. A new vision AI pipeline using Roboflow …
NVIDIA introduced intent-based security profiles in its Unified Fabric Manager for Quantum InfiniBand, enabling network administrators to configure multi-tenant fabric security with a single click. The three profiles — G…
Amazon Web Services launched a new intelligent document processing pipeline using Amazon Bedrock that offers both on-demand and batch inference options for extracting data from scanned PDFs and text documents. The soluti…
Data from the augmented reality game Pokémon Go was used to train artificial intelligence systems later adapted for military drone applications. The location and movement data collected from millions of players helped im…
Researchers trained a small transformer encoder to detect reward hacking in reinforcement learning trajectories by mapping them onto a unit sphere where embedding distance approximates reward-metadata signal differences.…
ERGO Hestia, a leading Polish insurer, reduced its time-to-market for pricing models by adopting Databricks' Lakebase and Mosaic AI Model Serving to unify data and model serving within a single lakehouse architecture. Th…
Cosmologists trained an AI neural network on simulations of the standard model of cosmology, then tested whether that pre-training would help or hinder the AI's investigations into new physics. The AI developed biases th…
A researcher developed an algorithm capable of computing optimal tokenizers in certain settings, solving a problem previously considered theoretically intractable. The finding mirrors results in Traveling Salesman Proble…
Roboflow released a workflow that automates medical device packaging inspection and UDI label verification by pairing its RF-DETR object detection model with Gemini 2.5 Pro. The system can detect damaged packaging, verif…