Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers at an undisclosed institution introduced Reliable Neural Collapse approximation (ReNC), a method for Open-World Test-Time Adaptation (OWTTA) that uses neural collapse as a structural prior to filter out-of-di…
A new arXiv preprint (2608.19801v1) finds that the choice of anomaly scoring function is critical for flow-matching-based unsupervised anomaly detection on tabular financial data, with trajectory-based Deviation and Reco…
Researchers introduced RecPFN, a prior-fitted network that brings in-context learning to sequential recommendation, achieving state-of-the-art zero-shot performance across eight public benchmarks. RecPFN is pretrained en…
A new arXiv paper (2608.19203v1) introduces Asymmetric Attention Heads (AAH), a framework that assigns different context window lengths to different attention heads in transformers, rather than giving every head the same…
Researchers propose MileGPO, a method for credit assignment in long-horizon agentic reinforcement learning that derives process-level credit from grouped on-policy rollouts via Milestone Discovery, Reliability-Calibrated…
A new study from arXiv (2608.19218v1) proposes a time-series retrieval framework that grounds multimodal large language models (MLLMs) in remaining useful life (RUL) estimation, improving prediction accuracy on the C-MAP…
Researchers introduced TT-Net, a conditional GAN architecture that replaces per-channel SVD denoising with a two-cut tensor-train decomposition to access cross-channel information, and in controlled comparisons it outper…
Researchers introduced QDOS (Quality-Diversity Offline Skill learning), a unified pipeline for offline-to-online reinforcement learning that uses an Advantage-Weighted Quality-Diversity pretraining objective to extract d…
A new arXiv preprint (2608.19535v1) proposes telemetry-informed adaptive compression for retrieval-augmented generation (RAG) on edge devices, based on experiments on the NVIDIA Jetson AGX Thor with Llama and Qwen genera…
SplatRAG 3, a retrieval system using local Gaussian-splat memory, achieved a SciFact (BEIR test, 300 claims, k=10) recall@10 of 0.8804 and nDCG@10 of 0.7624 on a store that also contains 5,000 Urban Dictionary rows, surp…
Moonshot AI detailed how it trained its Kimi K3 model, moving from architecture to a unified system through pretraining, SFT, reinforcement learning, and MOPD. The company trained the MoonViT-V2 vision encoder from scrat…
In the seventh part of a series on classic machine learning from an SRE perspective, an engineer explains that PCA can delete quiet but critical signals, such as a client health metric stuck at 2/10, because it prioritiz…
Australian start-up Cropify will commercially grade pulses such as lentils and chickpeas using artificial intelligence for the first time this harvest, a world-first that promises efficiency gains for the nation's $22bn …
Cosine similarity, the default metric in most RAG pipelines and vector databases, discards vector magnitude, which can carry critical semantic information such as confidence, specificity, or negation, leading to failures…
Waymo LLC, the autonomous driving unit of Alphabet Inc., revealed its custom 5nm application-specific integrated circuit (ASIC) and detailed its compute architecture, which has scaled 20x in eight years, in a blog post. …
Experiments on a Google Colab T4 with a 16 GB GPU show that SDXL's low-VRAM performance depends more on the granularity of runtime offloading than on the nominal memory budget, with finer-grained policies (group, leaf, s…
Researchers introduced HAWKEYE, an open-source framework that uses only 10 expert-written unit tests per GPU architecture to enable AI coding agents to generate hardware-aware kernels, approaching or surpassing expert-wr…
Hugging Face has built a hybrid search system for Papers with Code that combines PostgreSQL full-text search with pgvector dense embeddings, using reciprocal rank fusion (RRF) to merge results. The system maintains embed…
Local AI models can now answer 89% of everyday chat and reasoning queries as well as frontier cloud models, according to a November 2025 study by Stanford University and Together AI. The best local model's win/tie rate a…
A developer built a local embeddings pipeline using Ollama's nomic-embed-text model, ChromaDB, and Python, embedding local files and a live URL for semantic search. The pipeline encountered a context length error with a …