All Articles
16503 articles total
Convergence-Latency-Aware Adaptive Modulation and Resource Allocation in RIS-Assisted Wireless Federated Learning
This paper proposes a convergence-latency-aware resource allocation scheme for RIS-assisted wireless federated learning to optimize communication reliability and training speed.
An Isotropy-Preserving Spectral Cap for Muon: Theory and Three Case Studies
Researchers introduce a 'spectral cap' for the Muon optimizer to preserve isotropy and prevent training failure in large language models.
RPPNet: Perceptually-Grouped Rhythm-Pitch Primitives for Long-Term Structure Melody Generation via Boundary-Aware Modeling
RPPNet is introduced as a two-stage deep learning architecture that generates melodies based on psychological rhythm-pitch primitives for better long-term structure.
Physics-Aware Complex-Valued State Space Model with Scattering-Prior Feature Modulation for PolSAR Image Classification
CV-SSMNet is presented as a physics-aware complex-valued state-space network for improved PolSAR image classification using scattering-prior feature modulation.
OPIUM: Mitigating Steering Externalities and Over-Refusal via Dual Objective Latent Optimization
OPIUM is a training-free method designed to sanitize steering vectors in LLMs to reduce over-refusal and unintended externalities during inference.
Beyond Fail-to-Pass: Iterative Hardening of Co-Generated Bug Reproduction Tests and Fixes
CoHarden is a co-generation framework that iteratively hardens bug reproduction tests and fixes to improve automated program repair (APR) on SWE-bench.
Frameworkâs premium laptop is shipping with less RAM
Framework Laptop is adjusting memory configurations for its Laptop 13 Pro due to significant cost increases from its memory supplier.
Anatomy of a Sound Neural Reasoner: One-Shot Amortization, First-Pass Poisoning, and Search Inertness in Clue-Rich Completion
Researchers analyze the Lattice Deduction Transformer (LDT), finding it acts as a one-shot predictor rather than a true iterative reasoner in clue-rich Sudoku tasks.
PerfAgent: Profiler-Guided Iterative Refinement for Repository-Level Code Optimization
PerfAgent is introduced as a profiler-guided workflow that helps LLM agents perform repository-level code optimization by providing real hotspot feedback.
FedLSG: LLM-Enhanced Semantic Calibration for Federated Graph Backdoor Defense
FedLSG is a new framework that integrates LLMs into federated graph backdoor defense to improve resistance to poisoning attacks.
Reference-Free Evaluation of Reasoning in Open-Ended Question Answering
A new reference-free framework for auditing LLM reasoning traces is proposed, using NLI-hypergraphs to better evaluate reasoning in high-stakes domains.
SLPO: Scaling Latent Reasoning via a Surrogate Policy
Surrogate Latent Policy Optimization (SLPO) is introduced to enable outcome-reward RL for autoregressive latent reasoners, improving test-time scaling.
PhenSPINE: A Standardized Benchmark for Spine Pathology Diagnosis
PhenSPINE provides a standardized MRI dataset and benchmark for automating the diagnosis of spinal pathologies using deep learning.
Did Alice Do Wrong? Cross-Cultural Differences in Student Perceptions of Generative AI Use in University Computing Education
A study explores cross-cultural differences in how students from Canada and South Korea perceive the ethical use of generative AI in computing education.
Personalized Recommendation Tool Learning via Autonomous Language Agents
The PRTA framework uses an LLM as a central planner to interact with traditional recommendation models as tools for personalized ranking tasks.
An Automated Framework for Extracting Reachable Attack Chains from Cyber Threat Intelligence Reports
An automated framework is proposed to extract reachable attack chains from unstructured CTI reports using LLMs and Datalog-style rules.
Hybrid LLM-Guided Search for Quantum Reservoir Architecture Design
Introduces a simulator-based benchmark and a hybrid search policy combining LLMs with evolutionary algorithms to optimize Quantum Reservoir Computing architectures.
SynPre-FL: Synthetic data-driven pretraining integrated Federated Learning training framework
Proposes SynPre-FL, a framework that integrates synthetic EHR data generation with federated learning to improve clinical risk prediction under data scarcity.
D3VL: Understanding Driving Scenes from 3D Time Series Data and Video with Language Models
Presents D3VL, a multimodal LLM framework that integrates 2D video and 3D LiDAR data for enhanced autonomous driving scene understanding.
Trustworthy Privacy-Preserving Multimodal Federated Learning for Personalised Breast Cancer Prediction
Evaluates the use of multimodal federated learning for personalized breast cancer prediction, comparing its performance to centralized learning models.