2890 papers
This paper introduces TrustSLM, a trust-aware framework that aggregates outputs from multiple large language models using reliability metrics to detect hallucinations, identify the accuracy paradox of overconfident errors, and regulate responses through a controlled abstention policy for safer AI deployment.
This paper proposes YOLOv11-AHFE, a novel surface defect detection algorithm that integrates a dual-stream wavelet transform and a high-frequency enhancement module into the YOLOv11 architecture to preserve and recover critical high-frequency details lost during traditional downsampling, thereby significantly improving the detection of complex and small-scale defects on metal surfaces.
This paper demonstrates that adopting a minimax training objective, which optimizes network-redesign agents against a dynamically recomputed worst-case attack portfolio rather than a single fixed strategy, significantly closes the "Adaptive Degradation Gap" and produces topologies with superior worst-case robustness against reconnaissance-capable adversaries, regardless of the specific optimizer used.
This paper proposes AGHA-Net, a novel deep learning framework that integrates adaptive graph convolution, spatiotemporal gated recurrent units, and a dual-branch hybrid attention mechanism to significantly improve the accuracy and stability of short-term urban rail transit passenger flow prediction by effectively capturing dynamic spatial dependencies and complex temporal patterns.
This paper presents and evaluates an offline, hybrid pipeline for de-identifying Brazilian Portuguese clinical records using local open-weight language models and rule-based filters, demonstrating high recall and stability on modest hardware without requiring task-specific fine-tuning.
This paper proposes and evaluates an Agentic OODA Framework that leverages multi-agent systems with human-in-the-loop validation to overcome the variability and trust issues of standard LLMs, demonstrating significantly higher output consistency and reliability for transparent, safety-critical clinical decision support in stroke care.
This paper introduces MMANet, a privacy-preserving, video-only deep learning framework that leverages micro-macro temporal alignment to robustly estimate depression severity across diverse cultural datasets, achieving high accuracy without relying on noisy or intrusive audio and text modalities.
This paper proposes a molten mark classification method for electrical fire investigation that leverages a Swin transformer and a bi-directional feature pyramid network with directional cross-scale attention fusion to achieve superior accuracy and robustness against image degradation compared to existing CNN-based approaches.
This paper introduces FactLink, the first benchmark for the Fact-checking Link Role Classification task, which comprises two multilingual datasets of annotated links and demonstrates that while fine-tuned transformers excel in data-rich settings, few-shot Large Language Models offer superior robustness in low-resource and cross-domain scenarios.
This paper introduces OPERA, a Byzantine fault-tolerant execution model that extends concurrency across nodes via partial replication and inter-node parallelism, significantly reducing execution costs and boosting throughput in blockchains without compromising security or application semantics.