Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
A research team led by the Leibniz Institute for Food Systems Biology at the Technical University of Munich developed an AI method that predicts peptide bitterness and designs new bitter-tasting peptides from scratch, wi…
Research from the Bank for International Settlements shows that machine-learning systems can flag periods of likely financial-market dysfunction as far as 60 business days ahead, according to an analysis of how AI is bei…
A developer has introduced Jev, a decision model from TypeSafe that outputs typed decisions with calibrated probabilities rather than generated text, taking state and questions as input and returning answers such as bug/…
Vincent Sitzmann and co-authors submitted an arXiv paper on 17 June 2020 proposing sinusoidal representation networks, or Sirens, which use periodic activation functions for implicit neural representations. The authors r…
A developer published Rosetta M, a mock-up of a proposed APL2-tradition array language for machine learning, at sw-ml-study.github.io/m-poc, rendering ten small computations as "stones" with four faces each: the mathemat…
Convai Innovations released Laya, a free open-source model, on Hugging Face on September 18, 2026, three days after TypeSafe AI launched its $40 million seed-funded Jev model. Laya scored 0.766 accuracy on typed decision…
Rufus-Air is an open, reproducible post-training recipe built on GLM-4.5-Air-Base (106B-A12B) that runs as a serial pipeline of eight stages: SFT, Reasoning RL, Coding RL, Instruction-Following RL, General Agent, Coding …
Two research efforts report large cuts to the key-value cache used in transformer decoding. Grouped Value Attention (GVA) reduces persistent cache scalars by roughly 45–47% versus matched grouped-query attention while ke…
Black Forest Labs released FLUX 3 Action, a 7B open-weights World Action Model for robot control that ranks first on the RoboLab-120 leaderboard at 42.92% task success, ahead of NVIDIA's 16B Cosmos 3 Nano at 36.8% and π0…
A study posted to arXiv (2609.28703v1) proposes a hate speech detection framework combining DistilBERT embeddings, a Bi-LSTM, and an attention mechanism, with LIME used to explain predictions. The framework reached F1-sc…
ELF-REG, a method that scales Embedded Language Flows (ELF) continuous diffusion language models to reasoning tasks, reached 55.96% pass@1 on GSM8K at 64 network function evaluations and 13.39% on MATH-500 and 22.56% on …
A new arXiv paper (2609.29131v1) proposes a three-level framework for tag-aware structured text translation that combines a hybrid data synthesis strategy (Hy-LST), multi-task supervised fine-tuning across four sub-tasks…
Researchers released BanglaKontho, a 20-hour single-speaker Bangla text-to-speech corpus of 7,050 segmented utterances at 24 kHz drawn from professional audiobook recordings, alongside a reusable Bangla text normalizer a…
A prospective study of weak signals in embedding space found that documents embedding-based topic models initially classify as noise can be predicted at publication time to become founding members of emerging topics, wit…
Researchers introduced EAGER, a reinforcement learning framework for generative event extraction that pairs fine-grained verifiable rewards with Schema-Contrastive Advantage Estimation to counter advantage collapse under…
Researchers introduced UltraBench 2, a comprehensive benchmark for evaluating vision foundation models on ultrasound image analysis, with wide anatomical and task coverage and a focus on standardization, reproducibility,…
M-plicits, a multiscale neural implicit surface framework from researchers publishing on arXiv (2609.28684v1), models surfaces as a residual sum of MLPs trained over nested neighborhoods, achieving the best mean Chamfer …
Researchers released Smol-VL-BLV, a 500M-parameter vision-language model for blind and low-vision users that combines teacher-student distillation with Group Relative Policy Optimization (GRPO) using a composite BLV rewa…
DeltaWAM, a new world-action model from the AIGeeksGroup, improves average success on the RoboTwin benchmark over Fast-WAM from 81.3% to 85.4% in the clean setting and from 75.8% to 83.9% under visual randomization, acco…
Researchers introduced CinematicVQA, a benchmark for film-grammar reasoning in large vision-language models that uses a Cinematic Scene Graph (CSG) to link filming techniques to their perceptual effects and narrative fun…