All Updates
Page 897 of 907
February 12, 2026
Stats Test Spots LLM Degradations
McNemar's test framework detects post-optimization LLM degradations via per-sample comparisons. Aggregates across benchmarks with controlled false positives. Flags 0.3% drops confidently.
Silence Boosts Collective Taste Judgment
Introduces Silence Routing framework for collective intelligence in taste domains using music preferences. Specifies when contributors should speak, report, or stay silent. Simulation shows accuracy gains over baselines only when silence is allowed.
SigLIP Boosts Multi-Label ECG Classification
Adapts SigLIP contrastive learning with a Jaccard-based sigmoid loss for multi-label ECG classification using real-world data. Incorporates medical knowledge and techniques like higher embedding dimensions and random cropping. Per-label analysis identifies prediction challenges across ECG findings.
Semantic Labels Enhance TPRA Retrieval
Explores semantic labeling for TPRA questionnaires using LLMs and hybrid SSSL. Compares direct labeling vs. clustering and propagation. Improves retrieval when labels are discriminative.
Self-Supervised SR Quality Assessor
Proposes no-reference IQA for real-world super-resolved images using content-free SSL. Pretrains multi-SR model representations via contrastive learning. Includes new SRMORSS dataset for pretext training.
ScratchWorld Tests GUI Agents
Introduces ScratchWorld benchmark with 83 tasks for multimodal GUI agents in Scratch. Uses primitive/composite modes and execution-based evaluation. Exposes reasoning-acting gaps in state-of-the-art agents.
Safety Alignment for Omni-Modal LLMs
OmniSteer addresses cross-modality vulnerabilities in OLLMs using AdvBench-Omni dataset and modality-semantics decoupling. Uncovers mid-layer dissolution and extracts golden refusal vector via SVD. Boosts refusal rate to 91.2% while preserving capabilities.
SAF Improves Parkinson's ECoG Prediction
Introduces first reproducible ECoG dataset from rat models for Parkinson's disease prediction. Swap-Adversarial Framework (SAF) uses channel swapping and domain-adversarial training to tackle inter-subject variability and HDLSS issues. Outperforms baselines in cross-subject, cross-session, and cross-dataset settings, generalizing to EEG.
RSHallu: Hallucination Eval for RS MLLMs
RSHallu studies hallucinations in remote-sensing MLLMs with a new taxonomy, benchmark, and dual-mode checker. Provides datasets for mitigation via training and plug-and-play strategies. Improves hallucination-free rates by up to 21% on RS tasks.
Robust Policy Optimization for Recommendations
DRPO tackles model collapse in off-policy generative recommendation via optimistic distributionally robust optimization. Proves hard filtering recovers high-quality data from noisy logs. Achieves SOTA on mixed-quality benchmarks.
RLCER Evolves CoT Rubrics
RLCER reinforces chain-of-thought via self-evolving rubrics without human labels. Outperforms outcome-centric RLVR on reasoning tasks. Rubrics boost inference as prompts.
Rewiring Sparsifies Efficient GNNs
Explores adaptive rewiring and sparsification for scalable GNNs using ErdΕs-RΓ©nyi models. Tested on power grid N-1 analysis with GCN/GIN. Balances sparsity for generalization via tuning and early stopping.
RealHD Dataset Detects AI Fake Images
RealHD offers 730k high-quality real and AI-generated images from advanced methods like text-to-image and inpainting. Addresses prior dataset flaws with diverse prompts, metadata, and masks. Includes lightweight noise entropy detection baseline with strong generalization.
Quantum ICO Merges Sensing and Computation
Proposes quantum scheme using indefinite causal order (ICO) for integrated sensing and computation on one state. Agent superposes observation-then-compute and compute-then-observation orders. Achieves low losses in magnetic navigation tasks.
Quadrupeds Cooperate for Super Jumps
Co-jump enables two quadrupeds to synchronize jumps up to 1.5m via MAPPO and curriculum, without communication. Achieves 144% height gain over solo robots using proprioception. Transfers from sim to hardware.
ΞΌpscaling Optimizes Model Warm Starts
Proposes principled upscaling for model widths inspired by ΞΌP, with theory guaranteeing equivalence to widened versions. Extends ΞΌTransfer for hyperparameter scaling, avoiding costly retuning at larger sizes. Applicable to diverse architectures and optimizers with infinite-width analysis.
ProtoGLAD Enables Interpretable Graph Anomalies
ProtoGLAD detects graph-level anomalies by contrasting with nearest normal prototype graphs discovered via point-set kernels. It iteratively clusters normal graphs for unsupervised detection. Provides human-interpretable explanations outperforming black-box methods.
Privacy Shield for Mobile GUI Agents
Framework anonymizes sensitive UI data with type-preserving placeholders for cloud-based GUI agents. Detects PII across screenshots, XML, and instructions via layered architecture. Achieves top privacy-utility trade-off on benchmarks.
Privacy-Aware XR Collaboration Framework
PRISM-XR integrates multimodal LLMs for XR collaboration while filtering sensitive data from XR headset frames on edge servers. It features lightweight registration and customizable content-sharing for efficient synchronization. Evaluations show 90% accuracy in user requests and strong privacy protection.
Policy-Car Swerving Absorbs Traffic Jams
Proposes SVDD-JAD strategy mimicking police swerving to suppress stop-and-go waves via slow-in/fast-out maneuvers. Analyzes five key parameters measurable with roadside detectors. SUMO simulations confirm no secondary waves triggered.