2786 papers
This paper introduces "temporal probing," a direct validity test using difficulty-matched, post-cutoff items to measure data contamination in Bengali LLM benchmarks, finding that current models on the BoolQ-bn dataset show no significant inflation from memorization despite the prevalence of static, translated test sets.
This paper evaluates "World 8," an experimental shared-kernel architecture that decouples logical actor identity from transient sessions to enforce persistent, auditable, and recoverable multi-agent governance, demonstrating through extensive trials that while generic hardening addresses common faults, the specific governance wrapper is uniquely required to prevent complex authorization and tampering vulnerabilities.
This paper presents WVAB, a host-assisted wireless assistive vision framework that integrates ESP32-CAM sensing, UDP transport, and YOLOv8n inference to achieve low-latency, risk-aware navigation with multilingual audio guidance, demonstrating a 78.6 ms GPU processing latency and 91.5% planner agreement in prototype testing while acknowledging limitations in clinical validation.
This paper introduces ChronosAttack, a novel adversarial technique that manipulates the timing of authentic tool responses in asynchronous LLM agents to alter decision outcomes without modifying the content, demonstrating that response order itself constitutes a significant attack surface across various models.
This paper empirically demonstrates that while increasing parameter scale improves safety alignment in English, open-weight models remain significantly more vulnerable to jailbreaks when processing low-resource languages like Urdu and Punjabi due to script tokenization fragmentation, revealing critical evaluation blind spots in current safety benchmarks.
This paper investigates the relationship between feature selection and feature attribution methods through a comprehensive survey and unified evaluation framework, aiming to clarify their similarities, differences, and unique strengths to bridge the gap between these two research fields.
This paper outlines a protocol for developing and validating a localized Large Language Model, fine-tuned on 3,000 historical anesthesia records from Istishari Hospital, to generate preoperative anesthesia plans that are evaluated against human expert standards for safety and accuracy in a retrospective "Clinical Turing Test."
This paper proposes a robust coverless image steganography framework that securely transmits data without altering the carrier image by matching secret bits to the most significant bits of sliding window segments using a sine chaotic map, thereby achieving high resistance to various distortions and steganalysis while maintaining perfect invisibility.
This paper introduces TempoGFormer, a novel Transformer-based model for remote photoplethysmography that leverages a gating mechanism, a Temporal-Spectral Fusion Head, and a Spatially-Overlapped Tokenization strategy to effectively mitigate environmental noise and motion artifacts, thereby achieving superior accuracy and efficiency in physiological measurement from facial videos across multiple public datasets.
This study utilizes an eye-tracking investigation of static VR environments to demonstrate that while prior VR experience does not affect a user's willingness to explore, it significantly enhances visual efficiency and reduces cognitive load, ultimately informing evidence-based guidelines for optimizing scene design through strategic element placement and hierarchy.