Machine Learning News

A neuro-symbolic framework combining generative syntax with AraBERT resolved structural ambiguity in Modern Standard Arabic determiner phrases with 96.88% accuracy, 95.92% macro-F1, 96.83% weighted F1, and 93.94% binary …

Writerslogic's systems placed 1st in source detection (0.85 macro F1) and 3rd in safety classification (0.66 macro F1) on the PAN at CLEF 2026 Reasoning Trajectory Detection task, where training was entirely mathematics …

SEDIMA, a persistent hierarchical insight memory for LLM-driven evolutionary search agents, improves average final performance by 5.5% on AlgoTune and 6.6% on ALE-Bench LITE under a fixed budget of 100 evaluated candidat…

Researchers adapted a full-utterance speech language model for simultaneous speech translation using prefix supervision derived from its own complete- and partial-waveform translations, requiring neither transcripts nor …

Researchers introduced Sigma, described as the first large-scale continuous diffusion language model at 3B and 8B parameters, built on steerable low-dimensional ODE/SDE latent trajectories and trained blockwise via likel…

WakeKV, a reactive KV-cache residency policy, improves miss rate over frozen head classification and destructive eviction at matched memory or budget, according to an arXiv paper (2610.02713v1). Across three models (1.5B…

A new benchmark called OLMo-Detect, built on the fully open OLMo 2 pipeline, finds that membership inference attacks on large language models remain weak, with the best unsupervised and supervised attacks both reaching o…

HyperThink, a text-to-parameter hypernetwork approach described in arXiv paper 2610.03039v1, amortizes long-form reasoning computation into a single query-conditioned parameter update so the adapted model generates a con…

Researchers introduced MARETopic, a training-free topic modeling framework that selects exactly K exemplar documents from a corpus via rank-based prototype selection after projecting embeddings onto a low-dimensional man…

A comparative study of zero-shot authorship attribution (ZS AA) found that the two-stage LISA framework, which combines candidate space reduction with embedding-dimension selection, achieved the strongest overall perform…

The Writerslogic team's Claude Sonnet 4 submission ranked first among sentence-level systems in Task 1.1 of the CLEF 2026 SimpleText shared task with a SARI of 47.43 and BLEU of 14.21, placing third on the combined Task …

Researchers introduced Lexicographic Multi-Objective On-Policy Distillation (LMOPD), a multi-teacher method that integrates reward-specialized policies under explicit priority orders, according to an arXiv paper (arXiv:2…

Researchers introduced Queen, a 4B-parameter chess-language model that plays at Grandmaster level and explains its moves, gaining over 900 Elo points (1782 to 2697) across seven iterations of training, according to the a…

A research team posted arXiv:2610.02462v1, a framework for predicting capability loss from LLM compression across pruning, quantization, and distillation, measured in mathematics, code generation, and question answering.…

A study of on-policy post-training paradigms in large language models finds that the on-policy parameter update direction underlies generalization, arguing prior work treated these update behaviors only as byproducts rat…

Researchers introduced the World Embedding Benchmark, a dataset of 8,000 controlled simulation cases drawn from 80 families spanning fluid mechanics, to measure how video representations encode physical information. The …

Construct's benchmark of 117 labelled cases across five production decisions found Cloudflare's open-weight Clef decision model reached the right outcome on 92.3% of calls using thresholds set for TypeSafe's Jev, versus …

Apple researchers published a paper in October 2026 titled "Negotiating Ontological Boundaries in User-Authored Personal Sensing Systems," reporting a week-long exploratory study in which participants used one of two Wiz…

Eqx-zoo, a library by xquantize, loads Hugging Face Hub checkpoints directly as plain Equinox modules in JAX and verifies each model against transformers, supporting Llama, Qwen2, Qwen3 and Qwen3-MoE for generation with …

A new study of OLMo3 and Apertus pre-training checkpoints finds that language models repeatedly and abruptly switch between pattern-matching and generalizable computation, a behavior the authors call "mode-hopping." On a…