Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Apple Machine Learning Research reported that a single coding agent with shell access, called Malena, posted a 62.5% any-medal rate on MLE-bench, beating the best external multi-agent harness, AiScientist, by 15.4 percen…
Vision-language-action (VLA) models face a timing gap between low-rate inference and high-rate robot execution, according to a study that characterizes the gap through end-to-end latency measurements of model inference a…
A physics-informed neural network (PINN) was benchmarked against a plain finite-difference solver in two experiments, one in 1D and one in 5D, with the winning method changing between the two dimensions, according to a T…
Sarvam AI released Saaras V4, an automatic speech recognition model pairing an audio encoder with a 3B hybrid state-space LLM decoder trained from scratch in-house, which the company says achieves state-of-the-art perfor…
A hackathon organizer outlines an evaluation protocol for testing AI prototypes before final pitches, using a 30-announcement illustrative dataset split into 18 team-visible and 12 held-out examples for a JSON field-extr…
AI startup BioinvestGPT correctly predicted the outcome of five of six high-profile drug trials, including the negative result for Novartis' del-desiran, which missed its late-stage goal in September and erased $30 billi…
A new research paper introduces a framework for interpreting the dynamic contextual tokens that Multimodal Diffusion Transformers (MM-DiTs) form as they repeatedly update text tokens through multimodal attention during g…
A new technical note explains that in a shallow homogeneous neural net, minimizing the L2 norm of the parameters is equivalent to minimizing the L1 norm of a measure over unit-norm neurons, a duality the author says is d…
AbstractPhil released Beatrix V3, a 376-million-parameter byte-level language model with a 256-token vocabulary and 4,096-byte context, trained on 64.4 billion bytes across two RTX 5090 cards in 295 hours to 0.9861 bits …
OpenAI published a paper dated September 24, 2026, titled "Complex Matrix Multiplication Below 2.258 and Rectangular Bounds," claiming to prove that the matrix multiplication exponent omega is below 2.258, a drop of abou…
A multi-view self-verification framework called MOTIVE cuts hallucination rates by more than 20% on VQAv2, GQA, and ScienceQA while staying within a 2× speed budget, according to the MOTIVE paper (arXiv:2610.07018). The …
Google released EmbeddingGemma 2, an open-weights, lightweight multimodal embedding model that unifies text and image inputs into a single embedding space, removing the need to coordinate separate visual and textual vect…
OpenAI posted a paper titled "The Quasi-Riemann Hypothesis" claiming that every Dirichlet L-function, including the Riemann zeta function, has no zeros in the half-plane where the real part of s exceeds 11/12, a result r…
NVIDIA's NeMo-DCR introduces bit-exact delta-compressed refit for agentic reinforcement learning at trillion-parameter scale, addressing the 87.5-minute transfer time for a full 1T checkpoint between two AWS regions. The…
Google DeepMind released EmbeddingGemma 2, an open, Apache 2.0-licensed multimodal embedding model that maps text, code, images, audio and video into a single shared vector space for unified search. The 740-million-param…
A paper submitted to arXiv on 5 October 2026 introduces a decoder-only sparse autoencoder (SAE) backdoor that induces attacker-chosen behavior when the modified SAE is inserted into the forward pass of an otherwise uncha…
A paper titled TRACE introduces rollout-guided quantization-aware training for FP4 reinforcement learning of mixture-of-experts (MoE) language models, targeting the computation and memory overhead of rollout generation i…
Researchers proposed RADC, a risk-aware dual caching method for test-time adaptation of vision-language models, detailed in arXiv paper 2610.06932v1. RADC adds a Semantic Foreground Cache that aggregates category-consist…
A new arXiv paper (2610.06931v1) derives (ε,δ)-PAC guarantees for model-based risk-sensitive Q-value iteration (MB-RS-QVI) in finite discounted Markov decision processes under recursive entropic risk preferences with ris…
An arXiv industry experience report (arXiv:2610.06890v1) documents three years of running an event-driven cloud infrastructure for continuous machine learning training in automotive manufacturing, reporting a 72-78% cost…