Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Tokens per second for local LLM generation is primarily a memory bandwidth number, not a measure of silicon cleverness, according to an analysis by an unnamed author. At batch size one, generation speed equals memory ban…
A user on a forum is seeking advice on integrating AI models, including Hugging Face tools, with mass email sending services such as iDealSMTP for tasks like email personalization, subject line generation, spam-risk dete…
Duke University researchers have developed Raygun, an AI tool that shrinks proteins while preserving function, potentially easing gene therapy delivery. The system generated 70,000 fluorescent protein variants, with the …
A senior enterprise business intelligence leader argues that AI failures often stem from inconsistent and fragmented data rather than model inadequacy, citing a sales forecasting initiative where unstandardized opportuni…
A neural network trained to predict the stiffness of perforated metal sheets achieved perfect accuracy on held-out test sets but failed to generalize across different hole geometries, performing worse than a trivial dens…
Baidu's vLLM Kunlun plugin, open-sourced on Dec 8, 2025, enables vLLM to run on Kunlun XPU hardware, supporting 20+ models including Qwen, Llama, DeepSeek, GLM, and Gemma4, with features like quantization, LoRA, and spec…
A machine-learning model that estimates 'brain age' from sleep EEG recordings can flag dementia risk years before memory symptoms appear, according to a study led by UC San Francisco and Beth Israel Deaconess Medical Cen…
A developer building an ad click prediction project discovered that software engineering challenges, such as project architecture, testing, and environment issues, proved more valuable than the machine learning model its…
SAP SE is prioritizing tabular AI models over generative AI chatbots to deliver business value from structured data in its enterprise resource planning systems, according to a member-only Medium story. The company argues…
PolyAI released Dialog-RSN-1, an audio-native dialog model that processes caller audio directly, integrating turn-taking, speech recognition, function calling, and response generation into a single model while keeping te…
A senior developer argues that brute-force vector search with NumPy can outperform vector databases for datasets up to about 1 million documents, citing benchmarks from an M4 MacBook Pro showing 79.7 queries per second w…
PyTorch's multiprocessing module provides a CUDA Inter-Process Communication (IPC) API that enables sharing model weights across multiple processes for inference, avoiding duplication in GPU VRAM. The technique, demonstr…
A new method called Functional Reconstruction improves draft-token acceptance in speculative decoding for multi-head latent attention (MLA) draft models, according to a paper on arXiv (2607.27269v1). The method optimizes…
Researchers propose SDO (Structure-Aware Data Organization), a plug-and-play framework that organizes mini-batch composition and sample exposure based on representation-space structure to improve efficiency in large lang…
Researchers introduced MeshFM, a feedforward framework that distills 2D features from visual foundation models into 3D, enabling rich 3D shape understanding without 3D annotations. The method, which uses a two-stage trai…
Researchers propose MedXplore, a unified framework for reliable and unbiased Generalized Category Discovery in medical imaging, achieving an average 8.5% gain in All accuracy over the strongest competing methods on multi…
Researchers introduced GoGoTB, an agentic framework for register transfer level (RTL) verification that achieves specification-grounded coverage closure. In tests on 8 RTL designs without human intervention, GoGoTB achie…
Researchers introduced LayerRAG-Bench, a cross-layer reliability benchmark for agentic retrieval-augmented generation systems, covering 8 enterprise domains, 240 tasks, 9 fault scenarios, 2 contract modes, and 38,880 liv…
Researchers introduced EvoPINN, an agentic framework that uses a large language model (LLM) to automatically discover executable algorithms for physics-informed neural networks (PINNs), reducing relative L2 error across …
Researchers introduced CAPA, a benchmark with 600 coding sessions across 60 user-ambiguity cells, to test whether coding assistants can use a user's resolved session history to resolve recurring personalized ambiguities …