4158 papers
The paper introduces LEDGER, an evaluation platform that enables companies to create deterministic, auditable, and cost-aware tests for their business AI systems using proprietary data, demonstrating how a self-hosted open-weight model can achieve frontier-level accuracy at a fraction of the cost of proprietary alternatives.
This paper reveals that large language models generate summaries with strongly sublinear growth relative to input length, showing that while prompt directives have minimal impact, fine-tuning on length-specific data is the primary lever for controlling summary size.
This study demonstrates that an AI agent's accuracy in reading complex interior millwork drawings is significantly improved not by generic instructions, but by providing specific, context-rich information such as tiled sheet views, PDF text anchors, and targeted domain facts, which collectively transformed unstructured drawings into reliable structured data.
This paper addresses the lack of standardized resources for the Minimum s-Club Cover problem by introducing a fully reproducible benchmark suite comprising 30 diverse graph instances, 4,800 verified optimization runs across eight algorithms, and comprehensive open-access data to facilitate rigorous comparative evaluation and open science.
This study presents and evaluates an offline-first, quantized AI maternal-health assistant for low-connectivity Android devices, demonstrating that the Gemma model delivers clinically acceptable responses with effective local retrieval capabilities despite moderate latency.
This paper proposes HIGWO-NSGAII, a hybrid multi-objective optimization algorithm that integrates an improved grey wolf optimizer, Q-learning-based adaptive scheduling, and local search mechanisms to effectively solve multi-robot task allocation and path planning problems with superior convergence, stability, and scalability compared to existing methods.
This paper proposes a response-decomposed residual filtering module integrated into LEDNet's dynamic-kernel interfaces to jointly enhance low-light images and remove blur, achieving significant improvements in structural similarity and perceptual quality on a test set while incurring a modest increase in parameters and latency without boosting PSNR.
TriMagic is a single-image 3D generation framework that mitigates Janus artifacts and view inconsistencies by integrating semantic-role multi-token conditioning with dynamic pseudo-multi-view supervision to achieve superior fidelity and semantic similarity across benchmark datasets.
This paper proposes the Power Grid Innovation Graph Representation Learning (PGI-GRL) framework, which integrates scientific publications and patent data into a heterogeneous graph to jointly learn semantic, structural, and evolutionary features, thereby outperforming existing methods in discovering hidden innovation relationships and predicting latent technological pathways within power grid networks.
The paper introduces MV-CortexGuard, a multi-view, risk-aware deep learning network that enhances the detection of subtle pediatric wrist fractures by leveraging bone-axis-aligned cortical representations, anatomy-constrained cross-view consistency, and uncertainty-driven risk-controlled abstention to achieve superior accuracy and reliability compared to existing models.