4158 papers
This study demonstrates that single-run comparisons of frozen-CLIP adaptation methods for facial expression recognition are highly unstable due to training-seed variance, with most reported performance differences failing to replicate across multiple seeds and hyperparameter variations.
This paper proposes PGENet, a spatiotemporal prior-enhanced network that integrates a Prior-Aware Gating Block and a Prior-guided Kinematic Attention Block into the YOLO architecture to significantly reduce false positives and data payloads for low-latency fire detection in bandwidth-constrained industrial edge environments.
This paper proposes SPH-DiffDance, a coarse-to-fine diffusion framework that combines a Semantic Prototype Reconstruction module for stable global choreography and a Hypergraph-Enhanced Pose Refinement module for coordinated local details, effectively addressing the challenges of generating long-sequence, semantically consistent, and structurally rational music-driven dances.
DualTrust is a high-performance, low-latency runtime system that secures autonomous AI agents in ambient intelligence ecosystems by replacing slow, costly LLM-based guardrails with an asymmetric Bayesian reputation model and cryptographic gating mechanisms to enable real-time, tamper-evident authorization and containment.
This paper proposes LLM-GRec, a robust graph-based recommendation framework that leverages granular-ball computing to transform noisy, discrete LLM-generated user interests into stable multi-granularity anchors, thereby effectively bridging the semantic gap between language models and collaborative filtering to improve recommendation performance, especially in sparse scenarios.
This paper presents a hierarchical computational rationality model that successfully predicts how drivers adapt their visual sampling and control strategies to manage simultaneous lateral and longitudinal uncertainties under limited visual information, demonstrating strong alignment with human simulator data.
This paper presents a comprehensive systematic review of SMS phishing (smishing) attacks and defenses, categorizing existing research into user perception, attack characterization, defense mechanisms, and datasets to propose new directions for mitigating this rapidly evolving cyber threat.
This paper proposes a novel framework that resolves semantic conflicts in decentralized e-health workflows by combining model-theoretic semantics, zero-knowledge proofs for privacy-preserving verification, and Decentralized Autonomous Organizations to enable secure, interoperable governance without exposing sensitive clinical data.
This study challenges the hypothesis that segmentation improves predictive maintenance by increasing statistical separability, demonstrating instead that it enhances classification accuracy by reducing temporal ambiguity and structuring process dynamics, as evidenced by Sliced Wasserstein Distance analysis showing reduced statistical distance between classes in segmented representations.
This paper introduces PROMISE-Net, a novel medical image segmentation framework featuring Prompt-Conditioned Channel Attention (PCCA) that enables deep, hierarchical integration of spatial prompts to achieve consistent performance improvements across diverse anatomical structures and imaging modalities.