AI/ML arXiv cs.AI

ALLUDE: A Unified Evaluation System for Configurable Attacks in Differentiable Environments

Presentation of ALLUDE, an open-source unified evaluation system for configuring and testing adversarial attacks against vision models in differentiable environments.

AI/ML arXiv cs.AI

DepthART: Scaling Foundation Monocular Depth to Tiny Models

Development of DepthART, a compact monocular depth estimation model designed for efficient on-device deployment on tiny hardware like Jetson Nano.

AI/ML arXiv cs.AI

Auto Research for Materials: Auditable AI-Scientist Workflows with Held-Out Transfer

Research on using closed-loop AI agents to discover reusable and transferable modeling changes for materials science predictions.

AI/ML arXiv cs.AI

Teach it to stop, not just to click

An analysis of the variance and reliability of computer-use RL agents, providing a library for better k-seed reporting in research.

AI/ML arXiv cs.AI

Noise-Robust Box-Supervised Infrared Small Target Detection via Physics-Inspired Soft Label Optimization

Introduction of HALO, a physics-inspired soft label optimization method for noise-robust small target detection in infrared imagery.

Other Hacker News

Juggling for Blind People

An article about juggling for blind people, likely focusing on accessibility or human interest.

Software Engineering Hacker News

A flaky test exposed a Redis client use-after-free

A report on how a flaky test identified a use-after-free vulnerability in a Redis client.

AI/ML Hacker News

Show HN: An MCP server that turns async-work practices into tools

A new Model Context Protocol (MCP) server that converts asynchronous work practices into usable tools.

AI/ML Hacker News

How an AI Anime Is Created

An exploration of the technical process behind creating AI-generated anime.

Other Hacker News

Ten Steps Towards Happiness

A piece offering ten steps toward achieving happiness, which is non-technical content.

Tech Business/VC TechCrunch

Dimension Capital’s $800M third fund shows the intersection of science and compute is booming

Dimension Capital has launched an $800M third fund focusing on the intersection of science and compute.

AI/ML arXiv cs.AI

Counterfactual Shapley Credit Assignment

Introduces Counterfactual Shapley Credit Assignment and the $\phi$-PPO method to improve reward attribution in Reinforcement Learning.

AI/ML arXiv cs.AI

Scalable Causal Imitation Learning

Proposes Causal SQIL and Causal IQ-Learn algorithms to improve imitation learning in high-dimensional, confounded environments.

AI/ML arXiv cs.AI

Alignment of a Total Automation Economy

Discusses the economic theory and alignment vulnerabilities of a totally automated economy without human production.

AI/ML arXiv cs.AI

WHALE: A Scalable Unified Model for Recommendation with Wukong-HSTU Architecture

Presents WHALE, a unified recommendation architecture combining Wukong and HSTU with custom Triton kernels for efficiency.

Other The Verge

Samsung Galaxy Unpacked July 2026: How to watch

Samsung is preparing for its July 2026 Galaxy Unpacked event, where it is expected to unveil new foldable devices, including a redesigned Galaxy Z Fold 8.

AI/ML arXiv cs.AI

A Multi-Agent System for 5G Throughput Prediction in Multi-Operator Urban Environments

Researchers propose a Tiered Multi-Agent System (TMAS) to improve 5G throughput prediction in urban environments, achieving high accuracy and low latency.

AI/ML arXiv cs.AI

Optimizing Clinical Trial Protocols Using EHR-Derived Heterogeneous Treatment Effects

A study demonstrates that using heterogeneous treatment effects (HTE) derived from EHR data can identify specific patient subgroups with distinct responses to heart failure treatment.

AI/ML arXiv cs.AI

Pediatric Bone Age Prediction Using Deep Learning

A deep learning approach using EfficientNet with Additive Attention is presented for the automated prediction of pediatric bone age from X-ray images.

AI/ML arXiv cs.AI

What Do They See? Interpreting Complex Road Scenarios Through the Eyes of Vision-Language-Action Models for Safe and Trustworthy Autonomous Vehicle Learning

The Counterfactual Vision Action Analysis (CVAA) framework is introduced to interpret the causal influence of objects on the planning behavior of autonomous vehicle models.