AI/ML arXiv cs.AI

PerceptionDLM: Parallel Region Perception with Multimodal Diffusion Language Models

PerceptionDLM is a multimodal diffusion language model that enables efficient parallel region perception and captioning in images.

AI/ML arXiv cs.AI

Review of Machine Learning Models for Solar Energetic Particle Prediction

A review of machine learning models and datasets used for predicting solar energetic particle events to safeguard space technology.

Cybersecurity arXiv cs.AI

GDGU: A Gradient Difference-based Graph Unlearning Method for Cyberattack Localization in Electric Vehicle Charging Networks

GDGU is a gradient difference-based graph unlearning method designed for privacy-compliant cyberattack localization in EV charging networks.

AI/ML arXiv cs.AI

Exploring Feature Extraction Technique Parameters for Acoustic Gunshot Classification

A systematic study on how feature extraction parameters significantly affect the accuracy of acoustic gunshot classification models.

AI/ML arXiv cs.AI

FlowFake: Liquid Networks for Audio Deepfake Detection

FlowFake uses a Liquid Time-Constant architecture to detect audio deepfakes with high efficiency and superior cross-dataset generalization.

AI/ML arXiv cs.AI

A BART-based approach with hierarchical strategy for Vietnamese abstractive multi-document summarization

A technical report on a BART-based hierarchical strategy for abstractive multi-document summarization of the Vietnamese language.

AI/ML arXiv cs.AI

IHBench: Evaluating Post-Interruption Recovery in Voice Agents with Structured Workflows

IHBench is a new benchmark for evaluating how voice agents recover after user interruptions in structured enterprise workflows.

AI/ML arXiv cs.AI

PrefSQA: Pairwise Preference Prediction for Speech Quality Assessment and the Critical Role of High Quality Datasets

PrefSQA is a preference-based prediction method for speech quality assessment that reduces labeling noise compared to traditional MOS scores.

Hardware/Chips arXiv cs.AI

Interpretable and Verifiable Hardware Generation with LLM-Driven Stepwise Refinement

A new hardware generation framework that uses LLMs and formal methods to convert design specifications into verifiable RTL programs, reducing hallucinations in chip design.

Cybersecurity arXiv cs.AI

Execution-bound advisory automation for agentic AI: a reproducible AIBOM-driven CSAF-VEX framework

A protocol-driven framework for agentic AI that binds SBOM and AIBOM artifacts to runtime telemetry to automate CSAF-VEX security advisories.

AI/ML arXiv cs.AI

VERITAS: Verifier-Guided Proof Search for Zero-Shot Formal Theorem Proving

VERITAS is a zero-shot framework for formal theorem proving that uses a critic-guided MCTS pass to utilize verifier signals for improved proof search.

Software Engineering arXiv cs.AI

JustDiag!: A Diagnostic Justification Engine for Accountable Root Cause Analysis

JustDiag is a diagnostic justification engine for root cause analysis that emphasizes process accountability over simple fluent answers in high-stakes operations.

AI/ML arXiv cs.AI

Playful Agentic Robot Learning

Introduces RATs (Robotics Agent Teams) that use self-directed play to build a reusable code skill library for improved robot learning and task execution.

AI/ML arXiv cs.AI

Scaling Generative Foundation Models for Chest Radiography with Rectified Flow Transformers

A billion-parameter generative foundation model for chest radiographs using Rectified Flow Transformers to improve clinical data diversity and diagnostic robustness.

Cybersecurity arXiv cs.AI

Secure Coding Drift in LLM-Assisted Post-Quantum Cryptography Development: A Gamified Fix

Analyzes 'Secure Coding Drift' in Post-Quantum Cryptography development due to LLM reliance and proposes a gamified framework to mitigate these risks.

AI/ML arXiv cs.AI

Can In-Context Learning Support Intrinsic Curiosity?

Research on whether in-context learning can support intrinsic curiosity in machine learning by using prediction errors to drive data collection policies.

AI/ML arXiv cs.AI

Concept Flow Models: Anchoring Concept-Based Reasoning with Hierarchical Bottlenecks

Concept Flow Models (CFMs) introduce a hierarchical bottleneck for concept-based reasoning to reduce information leakage and improve model interpretability.

AI/ML arXiv cs.AI

Techniques for Peak Memory Reduction for LoRA Fine-tuning of LLMs on Edge Devices

A suite of memory-reduction techniques for LoRA fine-tuning of LLMs on edge devices, achieving up to 28x reduction in peak memory.

Open Source Hacker News

Ten years of ClickHouse in open source

A discussion on the ten-year journey of ClickHouse as an open-source project.

AI/ML arXiv cs.AI

ProMUSE: Progressive Multi-modal Uncertainty-guided Staged Evidential Alzheimer Disease Classification

ProMUSE is a staged evidential network for Alzheimer's classification that adaptively determines when expensive MRI or PET imaging is needed, reducing costs while maintaining accuracy.