AI/ML arXiv cs.AI

At-Grok Is Not Converged:A Measurement-Validity Audit for Grokking Representation Metrics

A measurement-validity audit showing that representational compression in networks continues long after the accuracy transition during grokking, challenging existing metrics.

Hardware/Chips Hacker News

Book: RISC-V System-on-Chip Design

A book discussing the design of RISC-V based Systems-on-Chip (SoC).

AI/ML arXiv cs.AI

Security and Privacy in Agentic AI: Grand Challenges and Future Directions

A research paper outlining the security and privacy challenges associated with the increasing agency of AI agents.

AI/ML arXiv cs.AI

D2PO: Optimizing Diffusion Samplers via Dynamic Preference

Introduction of D2PO, a framework that optimizes diffusion samplers using dynamic preference optimization to improve perceptual quality.

AI/ML arXiv cs.AI

Deep Reinforcement Learning for Reliability Based Bi-Objective Portfolio Optimization

A deep reinforcement learning framework (MORP-DRL) designed for bi-objective portfolio optimization considering reliability and risk.

AI/ML arXiv cs.AI

Audio Sentiment Analysis via Distillation and Cross-Modal Integration of Generated Multilingual Transcripts

A multimodal approach to audio sentiment analysis using distilled cross-modal transformers and multilingual transcripts.

AI/ML arXiv cs.AI

PRoVeFL: Private Robust and Verifiable Aggregation in Federated Learning

PRoVeFL is a new federated learning framework that provides privacy-preserving, Byzantine-robust, and verifiable aggregation.

AI/ML arXiv cs.AI

STAGformer: A Spatio-temporal Agent Graph Transformer for Micro Mobility Demand Forecasting

STAGformer, a spatio-temporal agent graph transformer, is proposed for more efficient micro-mobility demand forecasting.

AI/ML arXiv cs.AI

WHERE to Generate Matters: Budget-Aware Synthetic Augmentation for Label Skewed Federated Learning

FedEAS is a budget-aware synthetic augmentation policy designed to reduce label skew in federated learning.

AI/ML arXiv cs.AI

Inertia-1: An Open Exploration of Wearable Motion Foundation Models

Inertia-1 is an open exploration and 'cookbook' for building foundation models for wearable motion sensing data.

AI/ML arXiv cs.AI

Overview of the NLPCC 2026 Shared Task 1: Difficulty-Aware Multilingual and Multimodal Medical Instructional Video Understanding Evaluation

Overview of the DA-MIVQA shared task for NLPCC 2026, focusing on difficulty-aware medical instructional video understanding.

Other Hacker News

Spider venom kills varroa mites without harming honeybees

Researchers found that spider venom can eliminate varroa mites without harming honeybees, providing a potential biological control method.

AI/ML Hacker News

What's slowing down the AI buildout

A discussion on the bottlenecks and challenges currently slowing down the expansion of AI infrastructure and deployment.

Other Hacker News

The Strange Locomotion of Spirocuta

An exploration into the unique and unusual locomotion patterns of Spirocuta, a genus of bacteria.

AI/ML arXiv cs.AI

LipSSD: Lipschitz-Constrained Single-Shot Detection for Adversarially Robust Object Detection

Introduces LipSSD, a Lipschitz-constrained Single Shot Detector designed to improve the adversarial robustness of object detection systems.

Cybersecurity arXiv cs.AI

When Agents Remember Too Much: Memory Poisoning Attacks on Large Language Model Agents

Introduces the GhostWriter attack, demonstrating how LLM agents with long-term memory can be poisoned, and proposes AM-Sentry as a mitigation strategy.

AI/ML arXiv cs.AI

Non-contact, Real-time, Heart-rate Measurement using Image Processing with Commodity Cameras and AI Agents

A system for real-time, non-contact heart rate measurement using commodity cameras and image processing AI agents.

AI/ML arXiv cs.AI

MiLSD: A Micro Line-Segment Detector for Resource-Constrained Devices

MiLSD is a micro line-segment detector designed specifically for resource-constrained MCU devices with a sub-megabyte memory budget.

AI/ML arXiv cs.AI

TriRoute: Unified Learned Routing for Joint Adaptive Attention, Experts, and KV-Cache Allocation

TriRoute proposes a unified learned routing system to jointly optimize attention, expert selection, and KV-cache allocation in language models.

AI/ML arXiv cs.AI

Do Counterfactually Fair Image Classifiers Satisfy Group Fairness? -- A Theoretical and Empirical Study

A theoretical and empirical study on the relationship between counterfactual and group fairness in image classifiers.