Human-AI collaboration framework for spacecraft power system health management in satellite mega-constellations.
Method for post-training multi-turn interactive tool-using agents via self-evolving synthetic data and verifiable-reward RL, enabling complex multi-step task execution.
UAT-LITE addresses miscalibration in neural NLP models by adding inference-time uncertainty awareness to transformer attention without changing training or storage costs.
Examines behavioral mechanisms underlying user trust in chatbots versus normative trust frameworks.
FinTexTS dataset pairs financial time-series with textual data using semantic-based multi-level pairing for joint analysis.
Timer-S1 is a 8.3B parameter mixture-of-experts time series foundation model with serial scaling across architecture, data, and training.
Framework combining temporal graph attention networks with LLM explanations for supply chain risk prediction and interpretation.
Dynamic vehicle routing optimization for on-demand transit with advance request confirmation and real-time constraints.
Optimization-based camera calibration pipeline for sports field registration in broadcast videos using points and lines.
GNN-driven intrinsic rewards improve cooperation in decentralized multi-agent reinforcement learning with heterogeneous agents.
Sparse Variational Student-t Processes enable scalable Gaussian process alternatives for heavy-tailed data modeling.
Robust neural network training method addressing gradient path issues in quantization and sparsification for ultra-low precision regimes.
DRUPI reduces datasets using privileged information beyond input-label pairs, improving dataset condensation efficiency.
MKE-Coder applies multi-axial knowledge and evidence verification for automatic ICD coding in Chinese electronic medical records.
LLM-Advisor benchmarks LLMs for cost-efficient multi-terrain path planning in robot navigation tasks.
GateLens is an LLM agent for automotive software analytics that enhances reasoning on structured tabular data with safety-critical decision support.
Consequentialist critique of binary classification evaluation metrics, advocating for proper scoring rules over threshold-dependent metrics.
MCP Bridge provides a lightweight RESTful proxy for Model Context Protocol servers, enabling LLM tool integration on resource-constrained devices.
Stepwise Guided Policy Optimization improves GRPO for LLM reasoning by addressing the all-negative-sample group failure in reinforcement learning.
UltraEdit enables efficient lifelong knowledge editing in LLMs without training, preserving existing capabilities while updating with new information.
SATURN uses SAT-based reinforcement learning to improve LLM reasoning capabilities without heavy human annotation or expensive data synthesis.
Meta-learning approach to rate time series data quality using LLM judgments, extending to diverse domains beyond single-domain methods.
CORA method uses cooperative game theory to solve credit assignment in multi-agent reinforcement learning through coalitional advantage allocation.
Unified framework for multivariate time series forecasting handling inter-channel dependencies, sampling asynchrony, and missing values.
Supervised contrastive learning approach for low-resource language identification in multilingual LLM pretraining corpus curation.
OPENXRD benchmark with 217 expert-curated questions evaluates LLMs and MLLMs on crystallography X-ray diffraction question answering.
Theoretical model of mathematical problem-solving as belief-update loop combining prior knowledge, search, and information extraction.
Framework uses pretrained world models for robot policy learning across different embodiments by leveraging visual motion similarities.
LLM-agent framework simulates opinion evolution to study media influence on cross-border US-China attitudes and model bias sources.
Source-free domain adaptation method for facial expression recognition using personalized feature translation without labeled target data.
EgoCross benchmark evaluates multimodal LLMs on egocentric video question answering with domain shift across different real-world scenarios.
Educational approach using AI-generated singing and virtual avatars to present course syllabi for improved student engagement.
TaoSR1 deploys LLMs directly for e-commerce query-product relevance prediction using chain-of-thought reasoning with error mitigation.
Framework for adaptive chain-of-thought compression in LLMs reduces computational costs while maintaining reasoning quality on software engineering tasks.
VSSFlow unified flow-matching framework for both video-to-sound and visual text-to-speech generation tasks.
VoiceBridge one-step latent bridge model for general speech restoration from diverse distortions at 48 kHz.
v-HUB benchmark for evaluating multimodal LLMs on humor understanding using non-verbal short videos.
Latent Speech-Text Transformer improves compute efficiency of auto-regressive speech-text models through latent representation compression.
NavSpace benchmark with 1,228 trajectory-instruction pairs evaluates spatial reasoning and perception capabilities of embodied navigation agents.
RECODE framework uses code generation and derendering for visual question answering on structured visuals like charts and diagrams.
REAP demonstrates expert pruning outperforms expert merging for compressing Mixture-of-Experts models on generative tasks.
RL-100 framework combines diffusion visuomotor policies with reinforcement learning for real-world robotic manipulation tasks using clipped PPO.
Reasoning framework using LLMs with permutation relative policy optimization for interpretable tabular prediction with structural priors.
Vision-language-action model (FALCON) incorporating 3D spatial foundation priors for improved grounding and generalization in real-world robotic tasks.
Framework for synthesizing hand manipulation sequences with language instructions using discrete human-object interaction representations.
Vectorized parallel algorithm for POMDP planning under partial observability for autonomous robots leveraging modern hardware parallelization.
Graph domain-incremental learning method for updating models across multiple graph domains using knowledge disentanglement and preservation.
Structured matrix scaling approach for post-hoc multi-class classifier calibration beyond standard temperature scaling.
Data valuation method for time series foundation models using in-context fine-tuning to efficiently assess training data quality.
Multi-round entity-level reasoning segmentation task for medical images using text prompts, enabling iterative dialogue-based medical image analysis.