Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers developed the Mother-Child AI Agent (MoChiAgent), an LLM-based clinical assistant, which uses its predictive engine MoChiFormer to forecast maternal and infant diseases from electronic health records, achievi…
Common Crawl released the first machine learning embeddings built from its Web Graph, publishing 128-dimensional vectors for 52.9 million web hosts learned from hyperlinks alone. The dataset, trained on the November 2025…
SwarmOS, backed by GPT-6 Astra, generated a Rust/CUDA fully sharded data-parallel trainer for Qwen3.8-27B that achieved over 11.5× higher throughput than the eager PyTorch FSDP2 baseline and 21.5% higher throughput than …
Engineer Celcilin C S introduced CellularFlow, a novel architecture that replaces dense feed-forward networks in transformers with addressable DNA memory banks to solve catastrophic forgetting. The design achieves zero-b…
Nyra Labs released CrisperWhisper 2.0, a multilingual speech-to-text model that transcribes speech verbatim or as intended, with word-level timings averaging 30 ms boundary error on read speech and 41 ms on conversationa…
CellularFlow, a continual-learning LLM architecture from developer celcilin, replaces dense feed-forward networks with Multi-Head Associative DNA Memory Banks and an Episodic Memory Slot Buffer, achieving 83.9% retention…
An engineer's blog post explains the core mathematics of neural networks, covering perceptrons, weighted sums, activation functions, hidden layers, softmax, and loss functions. The post emphasizes that nonlinear activati…
A community kernel patch by S. Duggan re-enables 16 factory-masked Compute Units on the AMD GFX1013 die of the bc250 crypto GPU, boosting median generation speed by 1.32× and median prefill speed by 1.50× across 11 LLM m…
H Company released NeoMME, a family of 260M and 800M single-tower multimodal encoders that process multilingual text and raw 32×32 image patches in one Transformer without a pretrained vision tower or causal decoder. On …
Developer Rich Geldreich released neural block textures, a GPU texture compression method that replaces the fixed hardware decoder with a small configurable MLP (roughly 250-1200 weights) and PVRTC1-style bilinear endpoi…
Researchers introduced the first method for translating text embeddings between vector spaces without paired data, encoders, or predefined matches, achieving high cosine similarity across model pairs with different archi…
Meta's Fundamental AI Research (FAIR) team, with the University of Oxford and University College London, introduced AI Research Preference Models (RPMs) that rank unexecuted machine learning experiments before GPU hours …
A developer has outlined a plan to embed neural block texture codecs directly into the Khronos KTX2 texture format, using KTX2 global data to store MLP inference weights of roughly 250-500 parameters (up to 1k). The appr…
An experimental project called AGIO, a cognitive framework for systems, has been released publicly, with its creator reporting that fine-tuning a Dolphin model on AGIO data improved GPQA scores from 60/198 at checkpoint …
Developer omert27 released Calibra, an open-source toolkit for detecting issues in robot-learning datasets, designed to catch corrupted or invalid data before training. The tool targets users of LeRobot and imitation lea…
A GitHub project called DLSS-NR-on-AMD, published by modder Daniel Blanco, has successfully run NVIDIA's DLSS 5 neural rendering on AMD Radeon RX 9000 (RDNA 4) and RX 7000 (RDNA 3) graphics cards, achieving roughly 33 FP…
Towards Data Science published a tutorial demonstrating Linear Discriminant Analysis (LDA) for dimensionality reduction in classification problems, applied to a real-estate dataset. The article explains how LDA can be us…
A critical examination of Information Field Theory (IFT) and its NIFTY implementation, as prompted to Claude Sonnet 4.6, concludes that the framework's core statistical structure reduces to Gaussian processes and hierarc…
Google Research and DeepMind released WeatherNext 3, a weather model that learns directly from real-time satellite data instead of traditional physics simulations, producing hourly forecasts at up to five-kilometer resol…
A developer argues that OpenAI's newly released GPT-6 model, Astra, which replaces much of its chain-of-thought reasoning with uninterpretable numeric 'neural space' reasoning, is not the safety threat many media comment…