Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers introduced Direct Relational Set-Risk Pruning (DRSR), a method that compresses long-horizon language-model agent histories by scoring the risk of deleting sets of history blocks rather than individual units. …
Signal2Symbol, a neuro-symbolic framework for explainable biosignal anomaly detection, converts ECG and EEG signals into symbolic sequences via a learned VQ-VAE codebook or a SAX baseline, then scores anomalies using rar…
Researchers introduced LWCal, a CPU-only post-hoc probability calibrator for tabular classifiers that down-weights calibration examples whose noisy labels conflict with the base model's held-out probability, requiring no…
An analysis of a frozen Terminal-Bench 3 / Frontier-Bench 0.1 production record covering 1,081 pull requests, 639 scored tasks, 28,801 trials and $105,933 in logged agent spend found that only 78 of 125 all-fail tasks su…
A study of hindsight relabeling in preference-conditioned multi-objective reinforcement learning found the technique degrades 19 of 36 algorithm-environment settings by as much as four standard deviations, improves only …
Researchers proposed COPE (Continual Optimization with Personalized embedding and self-Evaluation), a framework that assigns learnable personalized embeddings to each user and uses self-evaluation to generate proxy rewar…
Researchers introduced QUARTET, a graph transformer architecture that combines a Causal Random Walk (CRW) sampler built on recency-truncated Personalized PageRank (PPR) with a quad-branch cross-attention module drawing o…
Researchers introduced Polymer Benchmark 2026 (PolyBench26), an open dataset of nearly 250,000 polymer-property datapoints spanning eight physical properties and four evaluation tasks across homopolymers and alternating,…
Researchers developed two distributed recursive Gaussian process algorithms, ADMM-RGP and PDMM-RGP, for multi-output GP regression across multi-agent networks, according to arXiv paper 2609.26979v1. The algorithms were v…
Researchers introduced Wasserstein-Tilted Flow Maps (WTF), a simulation-free reinforcement learning algorithm for fine-tuning flow-based generative models that achieves higher reward with comparable or higher diversity t…
Researchers introduced COMED (Controlled Model Escalation for Multi-LLM Deliberation), a post-anchor controller that selectively escalates queries to peer models instead of routing once or invoking peers on every query, …
A study on arXiv (2609.26926v1) found that having experts label cases of cross-LLM disagreement with rationales produced the highest LLM-labeling accuracy at 64.9% against expert labels, beating an expert-revised codeboo…
LexLattice, an extractive summarizer that models a legal act's hierarchy as a two-dimensional semantic lattice and consolidates over it with a masked 2D neural cellular automata, attained state-of-the-art ROUGE across al…
Researchers proposed Tempered Evidence Fusion (TEF), a training-free decision-fusion rule that weights each sentence's log-odds by its normalized information gain, for measuring value orientations in long social media po…
Researchers introduced RECAP (REdundancy-aware Credit Assignment via Propagation), a method that assigns step-level credit in large reasoning models using an LLM-annotated semantic dependency graph and gold-answer log-li…
Researchers introduced UGTPhon, described as the first grapheme-to-phoneme (G2P) benchmark for user-generated text in English, Vietnamese, and Korean, alongside an inference-grounded taxonomy for fine-grained diagnosis. …
Researchers introduced LOCKR, a hidden-state trajectory-guided planner that detects and repairs "stable-but-wrong lock-in" in diffusion language models, according to the arXiv paper 2609.27220v1. Across two diffusion lan…
A new arXiv paper (2609.27233v1) introduces a method for distilling sequential computation in Transformer language models by replacing spans of input tokens with collapsed surrogate embeddings generated on the fly by a l…
ChipMEM, a verification-grounded memory layer for LLM-based EDA agents described in arXiv paper 2609.27067v1, produced equivalence-passing outputs on 39 of 54 scored designs on RTLRewriter-Bench versus 35 of 54 without m…
Researchers developed resource-adaptive stochastic gradient descent (RASGD), an algorithm for stochastic online linear programming that requires O(m) operations for m resources per arrival and achieves O(log T) expected …