AI/ML arXiv cs.AI

Searching for Robust Augmentations to Improve Out-of-Domain Generalization in Dermoscopic Skin Cancer Classification

Research on using robust data augmentations to improve the out-of-domain generalization of dermoscopic skin cancer classifiers.

Open Source arXiv cs.AI

MediaWiki Code2Code Search: Neural Retrieval for the Semantic Discovery of Open-Source Software Entities

MediaWiki Code2Code Search is a neural retrieval system for semantic code discovery across MediaWiki repositories using FAISS IVF-PQ indexing.

Cybersecurity arXiv cs.AI

Physically Real-time Infrared Attack against Optical Flow Estimation Networks

Researchers developed a real-time physical attack on Optical Flow Estimation Networks using infrared lights to compromise autonomous driving and motion detection systems.

AI/ML arXiv cs.AI

Constitutional Midtraining: Content Presence Drives Alignment Gains

A study introduces constitutional midtraining, integrating principled values-based content into the training process of 120B scale models to improve alignment durability.

Cybersecurity arXiv cs.AI

Graph Is the Verifier: Agentic Reinforcement Learning for Interprocedural Vulnerability Detection

VulAgentRL is a new agentic reinforcement learning framework that uses Code Property Graphs to detect interprocedural vulnerabilities in software.

AI/ML arXiv cs.AI

Scientific Knowledge Discovery in the Age of Large Language Models

A survey paper examines the application of generative LLMs for scientific knowledge discovery, specifically literature retrieval and screening.

AI/ML arXiv cs.AI

Efficient Heteroscedastic Bayesian Optimization for Risk-Aware AutoRL

ERAHBO is proposed as a risk-aware Bayesian Optimization method to efficiently tune hyperparameters in stochastic reinforcement learning environments.

AI/ML arXiv cs.AI

MPEcho: A Melody and Phoneme-Aware Generative Framework for Controllable Cover Song Generation

MPEcho is a generative framework for controllable cover song generation that uses a new transcription model, Phonsa, to improve lyric accuracy.

AI/ML arXiv cs.AI

Automated Multilabel Mpox Research Classification with Explainable Transformer Models

A research project uses BERT and SHAP to automate the multilabel classification of Mpox research articles for better public health organization.

AI/ML arXiv cs.AI

FARI: Robust One-Step Inversion for Watermarking in Diffusion Models

FARI introduces a one-step inversion framework for faster and more robust watermarking in diffusion models via adversarial LoRA fine-tuning.

AI/ML arXiv cs.AI

Dual Inversion for Text-to-Image Diffusion Models: From Both Prompt and Noise Perspectives

Dualin is a two-stage method for text-to-image diffusion models that recovers both semantic prompts and latent noise to improve image fidelity and editing.

AI/ML arXiv cs.AI

Zero-Shot Face-to-Speech Synthesis via Latent Space Adaptation of a Style-Diffusion TTS Model

A new Face-to-Speech framework predicts a plausible voice from a static image by adapting a Style-Diffusion TTS model's latent space.

Other Hacker News

IMAX vs. IMAX 70mm: The difference between these two cinema formats

A discussion on the technical and visual differences between IMAX and IMAX 70mm cinema formats.

Tech Business/VC The Verge

Sony pushes forward with ditching discs, despite backlash

Sony is proceeding with the removal of physical game discs for PlayStation, citing the digitalization of content despite consumer backlash.

AI/ML arXiv cs.AI

Living-Harness Is an Interactive-Agent Evolver

Introduction of Living-Harness, a self-evolving agent harness that improves LLM agent reliability by converting trajectories into procedural knowledge.

AI/ML arXiv cs.AI

WhisperRec: Latent Reasoning for Efficient Foundation Recommendation Models

WhisperRec proposes a latent reasoning framework for foundation recommendation models that compresses Chain-of-Thought into latent tokens to reduce inference latency.

AI/ML arXiv cs.AI

Understanding Context Sampling in TabPFN on Small Tabular Datasets

A study on TabPFN, analyzing how context sampling and size impact prediction stability and accuracy on small tabular datasets.

Cybersecurity arXiv cs.AI

Guarding Organizations Against Malware Risk: A Novel Graph-Based Malware Detection Method

MalGuard is presented as a graph-based malware detection method that identifies operational roles to capture program behaviors and reduce risk.

AI/ML arXiv cs.AI

Filesystem-Based Memory for LLM Agents: Organization, Evolution, and Sustainability

A systematic exploration of using filesystems as long-term memory for LLM agents, finding that while organization reduces retrieval costs, it doesn't necessarily improve answers.

Cybersecurity arXiv cs.AI

Borrowed Strength: Best-of-N Search over a Code EncodingBreaks Self-Check Jailbreak Defenses

Research demonstrating how composing a code-completion encoding and best-of-N search can breach strong self-check jailbreak defenses like SAGE.