AI/ML arXiv cs.AI

Knowledge-Based Design Requirements for Generative Social Robots in Higher Education

Research on knowledge-based design requirements for generative social robots used as tutors in higher education.

Software Engineering Hacker News

Control the Ideas, Not the Code

A discussion on the philosophy of controlling ideas rather than the specific implementation of code.

Other Hacker News

The 'absolute magic' of Morse code that still connects people globally

An exploration of the enduring utility and 'magic' of Morse code in global communication.

Cybersecurity arXiv cs.AI

AutoGraphAD: Unsupervised network anomaly detection using Variational Graph Autoencoders

Introduces AutoGraphAD, an unsupervised anomaly detection method for networks using Heterogeneous Variational Graph Autoencoders.

AI/ML arXiv cs.AI

Point of Order: Action-Aware LLM Persona Modeling for Data-Grounded Civic Deliberation

Presents a pipeline for converting Zoom recordings into speaker-attributed transcripts to improve LLM persona modeling for civic deliberation simulations.

Hardware/Chips arXiv cs.AI

RIS-Assisted Downlink Pinching-Antenna Systems: GNN-Enabled Optimization Approaches

Proposes a GNN-enabled optimization approach for RIS-assisted pinching-antenna systems to maximize sum rate and energy efficiency in wireless communications.

AI/ML arXiv cs.AI

Data-Driven Learnability Transition of Measurement-Induced Entanglement

Uses neural networks in a self-supervised manner to detect measurement-induced entanglement in quantum systems, identifying a computational learnability transition.

AI/ML arXiv cs.AI

How to DP-fy Your Data: A Practical Guide to Generating Synthetic Data With Differential Privacy

A practical guide on generating synthetic data using Differential Privacy (DP) to protect user privacy while maintaining data utility for AI.

AI/ML arXiv cs.AI

ReinforceGen: Hybrid Skill Policies with Automated Data Generation and Reinforcement Learning

Introduces ReinforceGen, a hybrid system combining imitation learning and reinforcement learning for long-horizon robotic manipulation.

AI/ML arXiv cs.AI

Transition Matching Distillation for Fast Video Generation

Presents Transition Matching Distillation (TMD) to distill video diffusion models into efficient few-step generators for faster video generation.

AI/ML arXiv cs.AI

Principles of Lipschitz continuity in neural networks

A theoretical examination of Lipschitz continuity in neural networks and its role in robustness, generalization, and frequency signal propagation.

Software Engineering Hacker News

Show HN: DOM-docx – HTML to native, editable Word docs (MIT)

DOM-docx is a tool that converts HTML content into native, editable Microsoft Word documents under the MIT license.

Cybersecurity Hacker News

Leak of San Francisco Police Drone Footage Exposes Reality of Urban Surveillance

Leaked drone footage from the San Francisco Police Department reveals the extent of urban surveillance practices.

Tech Business/VC TechCrunch

Uber’s robotaxi lobbying effort puts it on a collision course with Waymo

Uber and Waymo are engaged in conflicting lobbying efforts in Washington D.C. regarding robotaxi regulations.

AI/ML arXiv cs.AI

Single-Frame Point-Pixel Registration via Supervised Cross-Modal Feature Matching

A new detector-free framework for single-frame point-pixel registration improves LiDAR and camera matching in autonomous driving.

AI/ML arXiv cs.AI

GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs

GrAInS is introduced as a gradient-based attribution method for inference-time steering of LLMs and VLMs to improve alignment and reduce hallucinations.

AI/ML arXiv cs.AI

Evaluating Retrieval-Augmented Generation vs. Long-Context Input for Clinical Reasoning over EHRs

Research indicates that Retrieval-Augmented Generation (RAG) is more token-efficient and effective than long-context prompting for clinical reasoning over EHRs.

AI/ML arXiv cs.AI

REAL: REtrieval-reAsoning and Logic-constructed Attention Behaviors for Long-Context KV Cache Compression

REAL is a new KV cache eviction method that uses multi-behavior analysis to significantly compress memory requirements for long-context LLMs.

AI/ML arXiv cs.AI

Contrastive Weak-to-strong Generalization

Contrastive Weak-to-Strong Generalization (ConG) leverages contrastive decoding to improve the transfer of capabilities from weak to strong LLMs.

AI/ML arXiv cs.AI

Explaining Human Choice Probabilities with Simple Vector Representations

A study uses simple vector representations to explain human choice probabilities in stochastic hide-and-seek tasks.