Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers have released Tpo-Torch, a PyTorch implementation of Target Policy Optimization (TPO), a reinforcement learning from human feedback (RLHF) algorithm that simplifies the standard PPO approach by eliminating th…
Nathan Lambert has released a course on RLHF and post-training for large language models, aimed at early AI PhD or master's students but accessible to anyone with basic machine learning knowledge. The course covers ML fo…
Yelp has launched Training Orchestrator, a new internal framework that replaces individual team Spark training scripts with a configuration-driven, DAG-based execution model to unify machine learning model training and e…
A new implementation of the Fuzzy-Pattern Tsetlin Machine achieves 37 million MNIST predictions per second at 98% accuracy with 4.7 GB/s throughput on a desktop CPU, and demonstrates the first Tsetlin Machine–based text …
A developer built a self-learning video editing agent using Claude Skills that can automatically edit a 30-minute video in 10 minutes. The agent separates five core editing tasks into independent Skills and features a fe…
Xiaomi trained its Xiaomi-Robotics-1 model on more than 100,000 hours of motion data collected by people using camera-equipped handheld grippers, finding that adding data improved performance far more than increasing mod…
Rhoda AI is betting the future of robotics starts with internet video, not lab data, using its Direct Video Action model that learns physics from hundreds of millions of clips. The company's headquarters visit revealed h…
AI scheduling platforms are being deployed in cardiac care to optimize surgical schedules and prioritize patients based on clinical risk, not just wait time. These systems analyze historical data, surgeon performance, an…
The Reserve Bank of India on June 24 published draft Guidance on Regulatory Principles for Model Risk Management that would require covered financial institutions to apply a board-approved framework to every model, inclu…
A leaked diagram sheet for Qualcomm's unannounced SM8475 chipset, believed to be the Snapdragon 4 Gen 6 processor, suggests the chip will include a dedicated Hexagon DSP and AI processor for the first time in the Snapdra…
A researcher proposed ShinBay-UCTF, a conceptual framework for privacy-preserving continuous learning, episodic memory, and compressed knowledge representation, outlining a phased development plan starting with UCTF rese…
Cisco AI engineer Shridhar Shah built a tiny multi-agent system where an agent learns to hallucinate future frames of a predator-prey game and then practices inside its own dream. Shah found that a dream model only stays…
A developer built a minimal implementation of the Lewis signaling game, where two AI agents invent their own language from scratch without any shared vocabulary. The agents achieved ~97% accuracy in coordinating to ident…
Evo Design released proto-tools, an open-source Python package that provides a unified interface for computational biology and biological AI tools, eliminating dependency conflicts by running each tool in an automaticall…
A developer built a neural network training loop in five lines using JAX, demonstrating gradient descent from scratch and achieving a 32x speedup with JIT compilation on a MacBook CPU. The project verifies JAX's gradient…
A developer benchmarked PyTorch 2.13's FlexAttention on an M1 Max Mac, finding up to 7.83x speedup over standard SDPA for sparse attention with 32,768 tokens and a 256-token local window. FlexAttention, introduced as a p…
A new production-inspired multi-stage framework called RouteCost improves pre-order shipping cost estimation in e-commerce by decomposing the problem into time-aware demand forecasting, fee-card-informed baseline pricing…
Researchers propose SelKV, a training-free framework for compressing the key-value (KV) cache in large language models (LLMs) that uses a soft cosine gate to selectively merge or discard tokens and an attention-ratio com…
Researchers introduce masked diffusion language models (MDLMs) as steerable text-based world models for reinforcement learning, achieving up to 47% absolute gains over baselines in zero-shot transfer across three out-of-…
A new study from arXiv evaluates nine open-weight small language models between 135 million and 3 billion parameters on a 1,085-example, 16-topic benchmark, finding that Qwen Coder 3B achieves the highest base accuracy a…