The Archive

IVD-SSM state-space model for SemEval-2026 narrative similarity task isolating causal patterns without Transformer quadratic cost.

Multiple instance learning framework trains cancer pathology classifiers on patient-level labels without report-level annotations.

Controlled failure modes in language models trained with RL reveal distribution shift vulnerabilities and post-training generalization failures.

Clinical NLP task on TNM cancer staging using ClinicalBERT, BioBERT, and classical ML on TCGA pathology reports.

WorldBagel: unified multimodal framework for Vision-Language-Action-World modeling in robotic manipulation tasks.

ADP: adversarial dynamics priors for robust humanoid locomotion control via CoM and contact force regularization.

Best-of-Better-N: in-context learning method to improve inference-time LLM alignment via retrieved high-reward responses.

SkillOpt-Lite: minimal skill optimization pipeline for autonomous agents via zeroth-order optimization, grounded in Claude.

HiMe: hierarchical embodied memory framework decoupling Vision-Language-Action control into frequency-stratified executor and reasoning tiers.

TRIAGE: stage-aware evaluation framework for LLM-automated Graph-RAG systems, localizing failures in extraction and inference.

Continuous test-time training for LLM agents across multi-turn episodes, balancing adaptation with drift prevention.

Router-oracle gap decomposition: separating reproducible specialist advantage from single-draw label noise in LLM routing.

ASIG: fine-tuning LLMs for sequential information gathering via amortized Bayesian Experimental Design and information-gain rewards.

Human-in-the-loop framework learns individual causal models to personalize algorithmic recourse recommendations in high-stakes ML decisions.

Dynamic Security Control Compositor addresses compositional policy violations in multi-tool agent chains via Most Restrictive Set algorithm.

Resemble AI's DETECT-3B-Omni deepfake audio detector demonstrates demographic and content independence across 10K samples from 8 voice-cloning systems.

SemEval-2026 task entry using transformer ensembles and LLM annotations for dimensional aspect-based sentiment regression.

S-ICDF dataset provides Sionna-simulated RF interference data for ML-based jamming detection, classification, and direction finding.

CO-ALIGN mitigates bias in text-to-image diffusion via concept-graph alignment in text encoders and denoisers without semantic collapse.

End-to-end analytics framework extracts urban mobility and commercial insights from mobile location data for tourism and resource planning.

Simon Willison's June 2026 newsletter covers Claude Fable 5, GPT-5.6, GLM-5.2 open weights, and US export restrictions.

DP-Diffusion compresses high-dimensional data (images) with differential privacy guarantees via diffusion models and stochastic codes.

Framework evaluates agentic policy repair in hotel-pricing simulator using only region-level diagnostic feedback without per-state labels.

Taxonomy of policy learning objectives beyond regret minimization, focusing on statistically significant improvement over baselines with limited data.

Dataset of 15.6M scientific papers with automatically classified rhetorical sections (Introduction, Methods, Results, Discussion) from S2ORC.

Spectral shape-based metrics using Heavy-Tailed Self-Regularization theory to fingerprint and compare LLMs across architecture/scale/training differences.

Google DeepMind partners with A24 on unspecified research initiative; details on focus area and technical outcomes unavailable.

CueTrust benchmark reveals vision-language models encode outlet-identity credibility priors, vulnerable to source-override attacks independent of article content.

CaSPECT: causal spectral clustering method for discovering homogeneous subgroups via directed acyclic graphs and treatment effect estimation.

Causal patching analysis reveals two visual information pathways in VLMs: direct (image tokens) and text-mediated (via query tokens), task-dependent.