Cybersecurity arXiv cs.AI

SoK: Attack and Defense Landscape of Mobile On-device AI Systems

A comprehensive systematization of knowledge regarding the attack and defense landscape of mobile on-device AI (MoAI) systems.

AI/ML arXiv cs.AI

Enhancing Flow Matching with A Unified Guidance Framework for Efficient and Robust Speech Synthesis

A new unified guidance framework for Flow Matching (FM) that reduces inference latency and improves speaker similarity in speech synthesis.

AI/ML arXiv cs.AI

When AI meets quantum information: A comprehensive review

A review paper exploring the bidirectional relationship between artificial intelligence and quantum information, discussing AI's role in designing quantum systems and QI's impact on AI models.

AI/ML arXiv cs.AI

MEPA: Multi-Scale Representation Alignment for Visual Autoregressive Modeling with Mixture of Experts

Introduction of MEPA, a scale-aware token-routed Mixture of Experts (MoE) architecture for improving Visual AutoRegressive modeling in image generation.

AI/ML arXiv cs.AI

Learning to Compose: Revisiting Proxy Task Design for Zero-Shot Composed Image Retrieval

Presentation of FoCo, a framework for Zero-Shot Composed Image Retrieval that uses proxy tasks to better model semantic modifications.

AI/ML arXiv cs.AI

MalariAI: A Label-Resilient Decoupled Framework for Universal Cell Segmentation and Explainable Stage Classification in Dense Malaria Blood Smears

MalariAI is a decoupled framework for cell segmentation and stage classification in malaria blood smears, designed for use in resource-limited settings.

AI/ML arXiv cs.AI

Learning Generalizable Skill Policy with Data-Efficient Unsupervised RL

GenDa is a unified framework for unsupervised reinforcement learning that improves data efficiency and generalizability through skill relabeling and an Information Bottleneck.

AI/ML arXiv cs.AI

NeuroCogMap Reveals Cognitive Organization of Large Language Models

NeuroCogMap is a framework that maps the internal functional organization of LLMs to cognitive capabilities and relates them to human cortical responses.

Other Hacker News

Aerial Photographs (2017)

A link to a collection of aerial photographs from 2017.

AI/ML VentureBeat

Z.ai launches ZCode to challenge Cursor, Claude Code and GitHub Copilot in AI coding

Z.ai launches ZCode, an agentic development environment powered by the open-source GLM-5.2 model trained on Huawei silicon.

AI/ML arXiv cs.AI

Testing Frontier Large Language Models' Physics Literacy in Parallel Physical Worlds

Researchers introduce a four-stage diagnostic to test LLM physics literacy in parallel, counterfactual physical worlds to distinguish reasoning from recall.

AI/ML arXiv cs.AI

What's Hidden Matters: Identifying Planning-Critical Occluded Agents using Vision-Language Models

A new framework using Planning KL-divergence (PKL) allows VLMs to identify and reason about planning-critical occluded agents in autonomous driving.

AI/ML arXiv cs.AI

An LLM-Based Framework for Intent-Driven Network Topology Design

This research explores an LLM-based framework for generating resilient network topologies from natural language requirements using a constraint-driven pipeline.

AI/ML arXiv cs.AI

Learning When to Listen: Gated Affect Fusion for Human Motion Prediction

The Gated Affect Transformer (GAT) is introduced to dynamically regulate facial affect information for improved human motion prediction in real-world videos.

AI/ML arXiv cs.AI

Mapping the Evaluation Frontier: An Empirical Survey of the Bias-Reliability Tradeoff Across Eleven Evaluator-Agent Conditions

An empirical survey analyzing the bias-reliability tradeoff in LLM evaluation systems across eleven different evaluator-agent conditions.

AI/ML arXiv cs.AI

RetailSMV: Exocentric vs. Egocentric Adaptation of Foundation Video World Models in Retail

The RetailSMV dataset is introduced to study the adaptation of foundation video world models in retail environments using egocentric and exocentric views.

AI/ML arXiv cs.AI

K-Inverse-RFM: A Modified RFM that Bridges the Gap to Neural Networks for Data-Corrupted Mathematical Tasks

K-Inverse-RFM is proposed as a modified Recursive Feature Machine that closes the performance gap with neural networks on data-corrupted mathematical tasks.

AI/ML arXiv cs.AI

DiscoLoop: Looping Discrete Embeddings and Continuous Hidden States for Multi-hop Reasoning

DiscoLoop is a new looping architecture that uses both discrete embedding and continuous hidden-state channels to improve multi-hop reasoning in LLMs.

Hardware/Chips The Verge

Amazon has enough satellites to launch its Starlink competitor

Amazon Leo has deployed 396 satellites, putting it on track for commercial availability by mid-2026 to compete with SpaceX's Starlink.

AI/ML arXiv cs.AI

EgoSafetyBench: A Diagnostic Egocentric Video Benchmark for Evaluating Embodied VLMs as Runtime Safety Guards

Introduction of EgoSafetyBench, a diagnostic egocentric video benchmark designed to evaluate Vision-Language Models (VLMs) as runtime safety guards for embodied AI.