3020 papers
This paper presents an ultra-compact, two-stage on-device audio cascade that utilizes a lightweight 2D CNN for snore detection and an adapted Coordinate Attention mechanism for classifying sleep apnea events, achieving high accuracy with a 93.2% parameter reduction compared to baselines.
This paper proposes a systematic conceptual framework and algorithmic extensions for spatio-temporal cartographic generalization that integrate temporal characteristics to preserve meaningful change events while ensuring map legibility, demonstrated through applications in data classification and line simplification.
This paper proposes an optimized Monodepth2-based monocular vision measurement method that integrates semantic segmentation and fixed-weight spatial correction using prior geometric features to achieve sub-millimeter accuracy in construction scenarios, demonstrating over 96% error reduction compared to conventional approaches.
This paper proposes a lightweight adaptive YOLOv11 detector integrated with a closed-loop cybernetic control system to achieve high-speed, high-accuracy industrial defect perception and real-time sorting, significantly improving inference efficiency and control stability compared to conventional methods.
This paper presents a reproducible benchmark demonstrating that while off-the-shelf self-supervised vision embeddings like DINOv2 can effectively generate leads for surface re-identification in OSINT, they remain insufficient for unqualified identification due to significant performance degradation across scale variations and on smooth surfaces.
This study challenges the assumption of language-invariant behavior in multilingual LLMs by demonstrating through a controlled empirical analysis that semantic, sentiment, and safety-related responses vary significantly across English, Hindi, and French, revealing that behavioral divergence is real and category-dependent rather than strictly following linguistic-distance predictions.
This quasi-experimental study demonstrates that AI-assisted non-programmers produced code with higher maintainability and structural quality scores than programmers working without AI, with the quality of human–AI interaction emerging as a strong predictor of these outcomes.
This paper systematically evaluates Claude's performance across high-stakes domains and languages, revealing that while the model maintains high output stability, it suffers from "normative drift" where its reasoning processes degrade and self-audit mechanisms become circular, leading to a novel failure mode called the "fortification pattern" where defended outcomes replace sound reasoning.
This paper proposes an empirically grounded, eight-pillar "core-plus-context" ethical design framework for AI-driven mental health applications, derived from qualitative interviews with stakeholders in Switzerland and Nigeria, to operationalize universal ethical principles while addressing distinct regional implementation challenges.
This paper presents a transparent empirical study of a PPO-based Bitcoin trading agent that diagnoses and corrects critical training failures to achieve modest long-horizon outperformance over Buy-and-Hold, while simultaneously demonstrating that real-time limit order book data often yields a rational "no-trade" policy due to transaction costs outweighing micro-price movements, ultimately advocating for the value of reproducible negative-result reporting over polished success narratives.