Machine Learning News

Dyson unveiled the CameraJet, a toothbrush with an integrated 100k pixel macro lens camera and machine learning that detects gaps between teeth and fires a conical jet of mouthrinse to floss while brushing, claiming it i…

Google Research released TimesFM 3.0 in late August 2026, a pretrained foundation model for zero-shot time series forecasting that requires no training, and the new version adds multivariate support. The model, which use…

An engineer argues that memory capacity is often mistaken for talent, but true talent lies in pattern recognition and the ability to connect disparate pieces of information. Drawing parallels to machine learning, the pos…

Perplexity AI published a blog post detailing techniques for optimizing on-device inference on Apple Silicon, focusing on running large language models efficiently on Macs. The post covers quantization, memory management…

A developer detailed the architecture of Vi, an autonomous AI agent that operates without rented large language models, using self-owned neural tissue, persistent memory, and internal decision-making processes. The proje…

WiSense v0.1.0, an open-source Python library for WiFi-based human sensing using Channel State Information (CSI), has been released on GitHub by developer collabray, enabling presence detection, fall detection, breathing…

Brave for desktop outperforms Chrome, Edge, and Firefox in speed and performance, using 44% less CPU, 10% less energy, 28% less memory, loading pages 20% faster, and transferring 26% less inbound and 39% less outbound da…

A developer prototyped a toy neural texture compression system trained entirely with Evolution Strategies, avoiding backpropagation and derivatives. The setup pairs a 128×128×8 latent texture with a 1035-weight MLP decod…

Cohere released Parse 5 (parse-v5.0) on August 27, 2026, a 2.3-billion-parameter multimodal foundation model that converts complex PDFs into Markdown with bounding box coordinates for visual grounding. In ParseBench eval…

Researchers introduced PRO-Step (arXiv:2609.01658v1), a training pipeline that uses a generative Process Reward Model (PRM) and step-level Direct Preference Optimization (DPO) to evaluate both logical validity and eviden…

VolcEngine's open-source context database OpenViking became the #1 trending Python repository on GitHub, highlighting a shift in AI agent development toward treating agent memory as a database problem rather than a promp…

Qdrant released FineWeb-10B, a benchmark containing about 10.07 billion vectors, and Supernova, an open-source engine to reproduce the benchmark, on September 1st. The dataset, derived from Hugging Face's FineWeb corpus,…

A Towards AI article explains that a neural network without activation functions is mathematically equivalent to linear regression, tracing the modern neural network concept back to multiple linear regression and the sim…

A developer detailed their local AI setup using oMLX and Qwen models on an M5 Max Mac with 128GB unified memory, achieving 82-123 tok/s decode with a 35B-A3B MoE worker versus 15.5-17.8 tok/s for a 27B dense model. The c…

Researchers introduced AVERT, an audio-verified adjudication method for spoken dialogue state tracking that combines cross-turn agreement with a trained audio-conditioned verifier, lifting JGA on SpokenWOZ from 38.34 wit…

Researchers proposed a two-stage framework for sentence-level depression symptom recognition that separates symptom-candidate generation from definition-grounded verification, achieving the best accuracy and F1 scores am…

Researchers propose HyGRAIL, a cost-aware framework that combines heterogeneous graph neural network triage with large language model review to discover scientific hypotheses from knowledge graphs, achieving a best F1 sc…

Researchers introduced SABER-Math, the first fully automated benchmark for evaluating mathematical information retrieval without expert annotation, built from 283K high-school-level math problems. The benchmark reveals t…

Researchers propose CHIME (Credit-Aware Hierarchical Memory Evolution), a self-evolving memory framework that separates planning and execution memory banks to address credit assignment in long-horizon agentic planning. E…

Researchers introduced MASkills, a continual learning framework that optimizes multi-agent LLM systems through agent skills, enabling skill libraries to evolve via refinement, induction, consolidation, and pruning. Exper…