Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
TypeSafe AI launched Jev, a hosted "System One" decision model, on September 15, 2026, and Convai Innovations released Laya, an Apache 2.0 open-weight alternative, weeks later. Both skip text generation to return typed l…
Xiaomi released and open-sourced the MiMo-V2.6 series, including the native multimodal MiMo-V2.6-Pro and MiMo-V2.6-Flash models, plus MiMo-V2.6-Distill-Qwen-9B and reinforcement-learning research resources. Xiaomi said M…
NVIDIA published LatentMoE in January 2026, a mixture-of-experts architecture that performs expert arithmetic in a space four times narrower than the model's width and spends the saving on a larger expert pool. On a 95B-…
Developer thevoxium released bare-lm, a minimal autograd tensor library written in C, installable via a git clone and make build or prebuilt Linux and macOS tarballs. The library provides a Memory object with 256 MB perm…
Transformer inference requires a KV cache to cut generation complexity from O(n^2) on the feedforward network and O(n^3) on attention down to O(n) and O(n^2) respectively, according to Part 7 of the "How To Scale Your Mo…
KV caching is a core mechanism inside transformer inference that splits the process into two stages: prefill, which runs the user's prefix through the model and computes keys and values for each token, and decode, which …
A new benchmark called GameHorizon Suite was introduced to evaluate AI models on multi-horizon data and gameplay tasks, according to the source. The suite targets visual understanding, instruction decomposition, goal pla…
Researchers proposed CLIC (Code Learning for Identification and Comparison), a visual analytics approach that characterizes large language model coding behavior through token-frequency analysis, training an interpretable…
Researchers at IIIT-Delhi's Complex Systems Laboratory released RecipeDB2, a structured compilation of 128,942 recipes and 35,474 ingredients spanning 32 regions and 99 countries, according to a new arXiv paper (2609.220…
A study published on arXiv (2609.22119v1) found that smaller language models detect evaluation via prompt format sensitivity while larger models rely on higher-order reasoning, based on tests of Gemma 3 (1B, 4B, 12B), Ph…
A new arXiv paper (2609.22131v1) proposes Correlation-Aware Structured Pruning, a method that models cross-unit dependencies in the reconstruction error as a cardinality-constrained binary quadratic program to cut Large …
Researchers posted arXiv paper 2609.22107v1 proposing a generalized multimodal foundation model that handles arbitrary modality combinations and prediction tasks without task-specific adaptation. The model is trained on …
A two-step runtime perception correction strategy preserved safety in 73% of 45 adaptive cruise control scenarios where perception-based control using Yolo and LaneNet had produced safety violations, according to an arXi…
Machine learning models predicting treatment retention and premature discontinuation in medications for opioid use disorder (MOUD) can show subgroup-level performance gaps even when overall predictive performance appears…
Researchers proposed the Activity Chain Encoder (ACE), a self-supervised Transformer model that combines pre-trained urban embeddings with visit timing and duration to model the sequential organisation of stays in mobile…
Researchers proposed LE4Mob, an inductive, distance-aware, geography-derived location embedding framework for human mobility modelling, according to a new arXiv paper (arXiv:2609.22117v1). LE4Mob extends contrastive lang…
A hierarchical Bayesian optimization framework using Gaussian process meta-modeling improved search efficiency and robustness over conventional surrogate-based methods for complex System-of-Systems architecting, accordin…
A study of 10,000 GPT architectures across 13 HW-GPT-Bench devices found that cross-device rank correlation with an RTX3080 proxy ranges from 0.951 to 0.996 in latency SRCC, yet proxy-reuse violation risk under matched j…
Researchers introduced StationPDE, a station-oriented surface PDE learning model that forecasts future weather variables at multiple weather stations from historical surface observations, according to an arXiv paper (arX…
Researchers proposed Independent Token Sampling (ITS), a query-efficient framework for detecting whether diffusion large language models (dLLMs) were trained on sensitive data, according to an arXiv paper (arXiv:2609.221…