Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers propose ReTA, a reinforcement learning-based dynamic topology augmentation framework that selectively imports external knowledge graph nodes and edges into electronic health record (EHR) visit graphs, outperf…
Researchers introduced OutageDiT, a generative foundation model for power outage forecasting and scenario simulation, trained on outage and weather records across the United States. The model generates seven-day outage t…
A new arXiv preprint (2609.01924v1) reports that looped and depth-recurrent transformers form a global-workspace-like band of causally potent representations, but recurrence changes how that workspace can be accessed. Te…
Researchers introduced a computational model that encodes symbolic knowledge as mental programs combining natural language and source code, using LLM-guided Bayesian learning to reproduce human inductive learning and act…
Researchers introduced Belief-Calibrated Optimization (BCO), a method that writes an LLM agent's implicit belief about environment responses into a persistent in-context document, forming an explicit world model that is …
A new arXiv paper (2609.01873v1) formalizes an 'epistemic Sybil problem' in multi-agent AI systems, showing that adding more AI agents does not add independent evidence when reports share common ancestry. In controlled t…
Researchers introduced DocHop, a benchmark for integrated chart-context reasoning in document-style images, finding that the best multimodal large language model achieves only 62.83% accuracy compared to over 90% for hum…
Researchers introduced PRO-STEP, a step-level process reward optimization method for retrieval-augmented generation that trains a generative process reward model to evaluate both logical validity and evidential grounding…
Researchers introduced MemeCULT-1K, a benchmark of 1,000 South Asian memes in Bengali, English, and Hindi, and found that providing minimal cultural context improved multimodal model performance across all 13 tested visi…
Researchers propose Potential-Guided Policy Optimization (PGPO), a new reinforcement learning method for multi-turn agentic tasks that estimates empirical state potentials from anchor-state-group return statistics to der…
A new arXiv paper (2609.02264v1) introduces Codebook Agent, a method that compresses successful communication topologies for LLM multi-agent systems into a 16-entry codebook and uses a reward-weighted MLP to select a top…
DocuSearch, an offline multi-agent document intelligence system developed for production telecom network operations, achieves Precision@10 of 0.69, Recall@10 of 0.79, and a grounding rate of 89.6% on a telecom corpus, ou…
Grab, Southeast Asia's leading superapp, is hiring a Principal Machine Learning Engineer for its AI Platform (AIP) team in Singapore, an on-site role reporting to the AIP Head of Engineering. The position involves archit…
Indicate, a new open-source tool for transliterating 12+ Indic languages to and from English, was released on Hacker News, offering both a PyTorch-based local model and LLM backends with auto-detection of source scripts.…
NVIDIA researchers report that their Nemotron-3-Ultra-CC (550B-A55B) system scored 535.4 out of 600 on the IOI 2026 problem set, exceeding both the gold threshold of 361.12 and the top human score of 498.27, marking the …
A new paper titled 'Test-Time Training with KV Binding Is Secretly Linear Attention' reveals that test-time training (TTT) architectures with KV binding, often seen as online meta-learning, can be expressed as a form of …
Researchers from ETH Zürich, Reichman University, and The University of Tokyo have developed an inverse rendering method that reconstructs fuzzy geometry such as hair and fur using explicit line segments rasterized on a …
A developer demonstrates how to use Microsoft's DoWhy library and CausalML to analyze health habits with causal inference, moving beyond correlation to isolate the true effect of caffeine on sleep quality. The tutorial e…
Hugging Face released funes, a durable memory layer for coding agents that indexes and retrieves session traces locally, supporting Claude Code, Codex, pi, and Hermes. The tool, available as a single binary, provides rec…
Fine-tuning LiquidAI's LFM2.5-350M model with Group Relative Policy Optimization (GRPO) using the TRL library improved its IFStruct benchmark score from 22.6% to 29.7% in just 100 training steps, according to a Hugging F…