Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Google's Gemma 4 model uses a 68k-parameter probe layer that predicts decoding errors by reading hidden states, achieving a 0.79-0.88 AUROC on audio benchmarks despite being trained on zero audio data. The probe enables …
A developer has built an upgraded version of the classic ELIZA chatbot using a sparse autoencoder with a large latent memory layer, achieving logarithmic memory scaling and resistance to catastrophic forgetting. The mode…
A guide to machine learning fundamentals explains the training loop, supervised learning paradigms, and the bias-variance tradeoff, emphasizing that models are functions optimized to minimize error through proper data sp…
The Huggy Database, the largest open dataset of French horse racing, has been released on the Hugging Face Hub, containing 356,371 unique races from 1996 to 2026 with approximately 5 million runners and detailed payout i…
BrainStem, an open-source bio-inspired AI architecture using 12 digital neuromodulators for lifelong learning, has completed Stage A foundation work and legacy cleanup, with the current database containing 102,275 chunks…
Samsung Electronics engineers will present a reinforcement-learning agent that tunes SoC quality-of-service settings at the DAC 2026 Engineering Track, demonstrating that major semiconductor houses are building their own…
A new study systematically decomposing OpenEvolve and TTT-Discover search harnesses into their constituent components and evaluating 30 budget-matched harnesses across 12 model-problem pairs using more than 3.1 million L…
A developer replaced a 7-billion-parameter LLM with a tiny Go classifier for email triage, achieving sub-millisecond inference on CPU with no GPU. The system uses deterministic rules first, then a TF-IDF and logistic reg…
Shinsegae Department Store's AI-powered hyperpersonalization technology, developed jointly with Seoul National University, has been accepted for presentation at the International Conference on Machine Learning, marking t…
A new approach called Auto-Fill, which post-trains three specialist small language models (SLMs) for world knowledge, text-based reasoning, and code-based reasoning, achieves superior accuracy in predicting missing cell …
A new framework called LAARA (Layer-Aware Adaptive Rank Allocation) dynamically assigns adapter ranks per transformer layer using lightweight diagonal Fisher estimates, outperforming LoRA, AdaLoRA, DyLoRA, and Bitfit on …
Researchers propose Multi-Mask Diffusion Models (MultiMDM) for few-step language generation, preserving the masking structure to enable drafting and refinement. The model achieves effective few-step generation through a …
Researchers propose FluxGraph, a reactive variant of TubeletGraph, that uses SAM2's internal multi-mask disagreement as a lightweight trigger for transformation detection, achieving ~3.3× speedup on VOST while improving …
A new unified variational framework for deep weakly supervised image segmentation uses a simplex-constrained Potts model with a smooth perimeter regularizer to create a convex, smooth energy functional that serves as a t…
Researchers introduce HyenaND, a subquadratic, global, input-dependent operator that acts directly on the native geometry of multidimensional data through convolutions with implicitly parametrized global, input-dependent…
A new benchmark, Air Quality Arena (AQA), covers 6 pollutants over 3 years across 7 countries and 4 continents, with more than 14,000 station-pollutant series, and finds that time series foundation models (TSFMs) consist…
A new study from arXiv introduces AdaRoPE, a method that assigns learnable rotation frequencies and attention scaling factors to each attention head in Transformers, outperforming standard Rotary Position Embedding (RoPE…
Researchers propose LISA (Linear-Indexed Sparse Attention), a plug-and-play attention module that reduces inference complexity from O(n^2) to O(nM) for long-context reasoning models. Experiments on DeepSeek-distilled-Qwe…
A new study from researchers at arXiv shows that hybrid offline-online reinforcement learning (RL) for vision-language-action (VLA) models achieves out-of-distribution (OOD) performance close to standard RL while requiri…
Researchers introduce Tabula, a privacy-preserving foundation model for single-cell genomics that uses federated learning and explicitly models tabular data structure, along with Chiron, a decentralized AI agent platform…