Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers from Sungkyunkwan University, Samsung Research, and KAIST presented the Unlearning Depth Score (UDS), a mechanistic metric that quantifies how deeply a model has unlearned target knowledge by measuring how mu…
NumPy 2.5, released in June, rewrote the searchsorted function in C++ to make batched binary searches up to 25x faster, according to the Scientific Python blog post detailing the optimization. The speedup applies to arra…
A developer built TrailSense, an offline-first web app that identifies bird songs and plants entirely on-device using the Web Audio API and in-browser computer vision. The app runs real-time FFT spectrogram analysis at a…
A developer's Day 11 explainer walks through how embedding models convert text into dense vectors of roughly 768 to 1,536 numbers so that semantically similar passages cluster together, and demonstrates cosine similarity…
A developer built Hephaestus, a local-first, open-source multi-agent system that takes a plain-language model description and autonomously handles research, dataset ingestion, PyTorch code generation, sandboxed training,…
A developer built StrollerPulse, an open-source, audio-first terrain monitoring system that uses a smartphone's accelerometer and gyroscope to sample vertical ground reaction forces at 50 Hz and classify surfaces via Pri…
AWS rolled out Intelligent Adaptive Scaling (IAS) for EC2 on October 24, a machine-learning autoscaling feature that predicts workload spikes with 95% accuracy and adjusts resources in real time instead of using threshol…
A developer built SkyReader, a browser-based app that uses an ONNX image-classification model running via onnxruntime-web to identify cloud types from a single phone photo, cross-references live local weather, and return…
A developer built a benchmark testing whether a general-purpose LLM can infer fraud from transaction-graph context rendered as text, using the IEEE-CIS Fraud Detection dataset. Gemma 4 31B (google/gemma-4-31b) scored 55%…
A developer built Trailworthy, an offline-first progressive web app that turns spoken trail observations into cited accessibility reports against the Architectural Barriers Act Accessibility Standard. The app runs Whispe…
ArXiv will cap submissions at two per person per month starting October 2026, after monthly submissions doubled in two years and grew more than sixfold in the cs.AI category. The preprint server said a small fraction of …
Software developer Michael Wright released X_eTaL-ML, a GitHub repository of ten machine-learning demos written in the X_eTaL array language and runnable in a browser or from a command-line clone. The demos include a Tin…
Software developer Michael Wright released X_eTaL Saw #14, adding a `xetal lint` linter with three initial checks and a Blender extension that lets X_eTaL programs generate the Python Blender runs in the background. The …
A persistent megakernel prototype for running a 27-billion-parameter model on a single Mac finished 2.701% slower than optimized separate kernels, according to an account by Abhishek Gahlot, despite improving 1.045% over…
A developer built Bird Walk, an offline bird identifier that runs entirely on a laptop with Wi-Fi disabled, combining BirdNET for audio classification with a local Gemma 1B model for short field notes. The tool filters c…
Tamoghna Chattopadhyay and colleagues at the University of Southern California published MEMOIR-VLM on October 1st, a research model that classifies Alzheimer's disease from whatever combination of MRI, diffusion imaging…
A developer built SolSpot, an open-source anti-burnout app that uses Prior Labs' TabPFN regressor for zero-shot prediction of a personal 20-minute daylight window from hourly atmospheric vectors, then has Google's Gemma …
A physics undergraduate documented a step-by-step workflow for training a first neural network on Windows using WSL 2, Ubuntu, a Python virtual environment, and PyTorch. The guide walks through installing Ubuntu via `wsl…
Stanford PhD student Ziga Kovacic and collaborators introduced 4DCodeBench, a 200-scene benchmark testing whether coding agents can turn video of a physical event into executable graphics code reproducing geometry, appea…
Researchers in Prof. Michal Irani's lab at the Weizmann Institute of Science developed Brain-IT, an AI model that reconstructs viewed images from fMRI brain activity and needs only one hour of scans to learn to read a ne…