The Archive

Study evaluates robustness of 11 ML-based anomaly detectors for industrial control systems under training-time data contamination on SWaT.

IMNO neural operator exploits low-dimensional inertial manifolds for improved stability and accuracy in dissipative PDE prediction.

Controlled study shows AI assistance improves short-term task performance but degrades skill retention when assistance is removed, with cost variations affecting usage patterns.

Analysis of multi-agent LLM interaction effects finds communication reduces diversity and may not improve quality under equal computational budgets across different model families.

Local linear distillation method achieves interpretability in high-accuracy models like tabular foundations and gradient-boosted ensembles through local regression approximation.

Few-shot continual learning approach using adapters addresses catastrophic forgetting in malware detection systems for new threat variants with minimal labeled data.

Multimodal sensor and clinical assessment study predicts joint recovery outcomes for older adults after lower-limb fractures using 1,008 participant-days of monitoring data.

DreamerV3 world model learns energy-like scalar invariants that improve video prediction rollouts when latent states are projected toward conserved level sets.

EarthVerse benchmark evaluates scientific agents on 405 reproducible earth-system tasks across 19 natural hazard families, testing reasoning on heterogeneous multimodal evidence.

Review of AI-driven measurement in economics identifies three pipeline stages (discovery, construct definition, observation) and measurement validity issues when multiple plausible variables emerge.

MHER benchmark for historical entity reconciliation contains 396 name pairs from Mongol world across scripts and languages, addressing identity matching beyond string similarity.

Safety-direction penalty training mitigates reasoning-induced misalignment where chain-of-thought fine-tuning triggers harmful LLM behaviors through representation-space geometry.

SRPO framework enables LLMs to self-reflect on trajectories and synthesize errors into dense training signals for improved long-horizon reasoning.

Team Semiintelligencn's multi-modal retrieval pipeline for ACM RecSys 2026 music recommendation challenge using seven embedding spaces and RRF fusion.

Multi-objective optimization framework for sequential clinical classification balancing early intervention timing against monitoring cost and diagnostic accuracy.

Intention Distillation method improves Vision-Language-Action models by supervising shared semantic objectives of robot behavior beyond behavior cloning.

Investigation of weird generalization phenomenon where narrow fine-tuning causes broad behavioral changes; analyzes dataset features and evaluation sensitivity.

Nvidia worker indicted after Jensen Huang scolded Supermicro for AI server smuggling.

StrategyBench evaluates whether LLMs can induce explicit task rules from examples in few-shot learning, addressing sensitivity to example construction.

TimeCatch benchmark evaluates temporal consistency in vision-language models using frame-swapping and noise injection anomaly detection.

MetaCaster meta-learning framework enables few-shot training of lightweight time series forecasters for resource-constrained multimodal scenarios.

InjecMEM attack paradigm demonstrates memory injection vulnerability in LLM agent systems requiring only single interaction to steer responses.

ML framework for inverse design of 19x23 pixelated millimetre-wave patch antennas in 22-30 GHz band using CST simulations.

ProxyFormer reduces quadratic attention overhead in ultra-long-context LLMs via dual-stream proxy token architecture with bottom-up compression and top-down decompression.

Empirical evaluation of diversity-based active learning metric spaces (K-center, K-median) for cost-effective classifier training with minimal labeled data.

Influence functions adapted for label-free data attribution and error detection in ESA's Ariel space mission spectroscopy pipelines without ground truth.

Reward-free continual learning framework using latent-state world models for online robotic adaptation in space environments with hardware degradation.

llm-anthropic 0.27 updates Python SDK compatibility with anthropic v1.0.0, migrating from httpx to httpx2.

Space robots operate in extreme environments where hardware degradation can critically compromise traditional control strategies. While continual reinforcement learning offers a promising mechanism for online adaptation, it inherently requires access to a reward signal during deployment. However, precise reward computation in space is often infeasible due to the lack of external tracking systems and the overall complexity of the environment. To address the challenge of unobservable rewards, we introduce a reward-free continual learning framework that leverages latent-state world models. By pr...

Controlled evaluation of GPT-5.4 and gpt-oss-120b on knowledge-driven grammar engineering for Cantonese ParGram resources with varied prompting conditions.