AI/ML arXiv cs.AI

OmniPath: A Multi-Modal Agentic Framework for Auditing Wheelchair Accessibility

OmniPath is a multi-modal agentic framework that uses aerial LiDAR and OpenStreetMap data to audit wheelchair accessibility and identify physical hazards.

AI/ML arXiv cs.AI

T2D-Bench: Evidence-Gated Evaluation of LLM Outputs for Type 2 Diabetes Using a Multi-Layer Clinical-Lifestyle Knowledge Graph

T2D-Bench is a reproducible evaluation framework using a multi-layer clinical-lifestyle knowledge graph to verify LLM medical recommendations for type 2 diabetes.

AI/ML arXiv cs.AI

The Geometry Behind Diffusion and Flow Matching: Gradient Flows and Geodesics in Wasserstein Space

This paper provides a mathematical unification of diffusion models and flow matching by framing them as gradient flows and geodesics within Wasserstein space.

AI/ML arXiv cs.AI

An Introduction to Causal Reinforcement Learning

This work introduces Causal Reinforcement Learning (CRL), a unification of causal inference and RL to better handle counterfactual reasoning in agent environments.

AI/ML arXiv cs.AI

Data Scale, Not Latency, Shapes Cross-Lingual Encoder Transfer in Streaming ASR

Research on streaming ASR shows that multilingual initialization is highly beneficial in low-data regimes but becomes irrelevant as target-language data increases.

AI/ML arXiv cs.AI

Navigating User Behavior toward Personalized Multimodal Generation

NaviGen is a framework for personalized multimodal generation that translates user interaction history into executable instructions for image and video synthesis.

Tech Business/VC arXiv cs.AI

Exploring the relationship between human-centric AI and firm idiosyncratic risks

A study on Chinese listed firms suggests that human-centric AI strategies can reduce a firm's idiosyncratic financial risks.

AI/ML arXiv cs.AI

FlowR2A: Learning Reward-to-Action Distribution for Multimodal Driving Planning

FlowR2A uses a flow-matching decoder to learn reward-to-action distributions, improving the quality of multimodal driving planning proposals.

AI/ML arXiv cs.AI

SP-Mind: An Autonomous Reasoning Agent for Spatial Proteomics Analysis

SP-Mind is an autonomous AI agent that automates the end-to-end spatial proteomics analysis pipeline from tissue imaging to phenotype discovery.

Software Engineering Hacker News

Show HN: An ASCII 3D Rendering Engine

A showcase of a 3D rendering engine that outputs graphics using ASCII characters.

AI/ML arXiv cs.AI

Critique of Agent Model

A theoretical critique of AI 'agency', distinguishing between engineered workflows (agentic) and endogenous capabilities (agentive), and proposing the GIC architecture.

AI/ML arXiv cs.AI

Safe and Generalizable Hierarchical Multi-Agent RL via Constraint Manifold Control

A hierarchical multi-agent RL framework that uses constraint manifolds to provide theoretical safety guarantees in safety-critical applications.

AI/ML arXiv cs.AI

Reinforcement Learning Towards Broadly and Persistently Beneficial Models

Research demonstrating that RL focused on beneficial traits (truthfulness, fairness) in specific domains can generalize alignment to unrelated domains and resist adversarial prompting.

AI/ML arXiv cs.AI

Can Language Model Agents be Helpful Circuit Explainers in Mechanistic Interpretability?

Introduces HyVE, an agentic explainer for mechanistic interpretability, and AgenticInterpBench to evaluate how LLMs can explain transformer circuits.

AI/ML arXiv cs.AI

Breaking the Filter Bubble: A Semantic Pareto-DQN Framework for Multi-Objective Recommendation

A multi-objective RL framework using Pareto-DQN to balance user engagement with information diversity and fairness in recommender systems.

AI/ML arXiv cs.AI

Ensemble Feature Selection and Harris Hawks Optimization for Explainable Mental Health Risk Prediction in Female Sex Workers

A hybrid ML model combining ensemble feature selection and Harris Hawks optimization to predict mental health risks in female sex workers.

AI/ML arXiv cs.AI

Beyond Trajectory Imitation: Strategy-Guided Policy Optimization for LLM Reasoning

Proposes Strategy-Guided Policy Optimization (SGPO), a method for distilling reasoning strategies from strong LLMs to weaker ones, outperforming standard trajectory imitation.

AI/ML arXiv cs.AI

Exploring Academic Influence of Algorithms by Co-occurrence Network Based on Full-text of Academic Papers

A large-scale analysis of NLP algorithm influence using co-occurrence networks derived from the full text of academic papers.

AI/ML arXiv cs.AI

ReMMD: Realistic Multilingual Multi-Image Agentic Verification for Multimodal Misinformation Detection

Introduces ReMMD, a framework and benchmark for detecting multimodal misinformation using an agentic verification process.

Hardware/Chips Hacker News

Raspberry Pi Pico W as USB Wi-Fi Adapter

A discussion on utilizing the Raspberry Pi Pico W as a USB Wi-Fi adapter, focusing on hardware tinkering and connectivity.