AI/ML arXiv cs.AI

CausalGraphX: A Counterfactual Graph Neural Network Framework for Explainable Systemic Risk Assessment

Presents CausalGraphX, a framework combining GNNs and counterfactual reasoning for explainable systemic risk assessment in financial networks.

AI/ML arXiv cs.AI

Per-Token Fixed-Point Convergence in Depth-Recurrent Transformers

Analysis of depth-recurrent transformers showing per-token fixed-point convergence, enabling a training-free early-exit rule to reduce computation.

AI/ML arXiv cs.AI

Tactile: Giving Computer-Using Agents Hands and Feet

Introduces Tactile, an open-source tool layer that provides agents with a reliable, semantic interface for desktop automation beyond simple coordinates.

AI/ML arXiv cs.AI

Step-Level Preference Learning for Generative Agents in Social Simulations

Proposes step-level preference learning for generative agents in social simulations to improve fidelity and interaction quality.

AI/ML arXiv cs.AI

SAGA: Schema-Aware Grounding for Agentic Text-to-SPARQL Generation

Introduces SAGA, a training-free framework for schema-aware grounding in agentic Text-to-SPARQL generation to reduce 'type-blind' errors.

AI/ML arXiv cs.AI

Contextualized Evaluation of Vision Language Models through Dynamic, Multi-turn Interactions

Presents CEDI, a framework for the contextualized, multi-turn evaluation of Vision Language Models to better identify real-world hallucinations.

AI/ML Hacker News

What loss.backward() actually does

A discussion on the internal mechanics and mathematical operation of the loss.backward() function in deep learning frameworks.

AI/ML arXiv cs.AI

Automatic Hard Example Synthesis with Multi-Level Agentic Data Curation

Proposes an automated agentic red-teaming framework to synthesize difficult adversarial examples for improving the robustness of Multimodal LLMs.

AI/ML arXiv cs.AI

AI Agents Do Not Fail Alone:The Context Fails First

Presents ProofAgent-Harness, an open-source infrastructure for evaluating AI agent reliability by measuring context-engineering quality.

Other arXiv cs.AI

Measuring How Students Rely on Generative AI in Academic Writing: Development and Multi-Source Validation of the Generative AI Reliance Types Scale (GenAI-RTS)

Develops the GenAI-RTS scale to measure four types of student reliance on generative AI in academic writing.

AI/ML arXiv cs.AI

Tracing LLM Behavior to the Training Data with Empirical Next-Token Distributions

Explores 'data-centric mechanistic interpretability' by comparing LLM next-token distributions with empirical distributions from training data.

AI/ML arXiv cs.AI

Traccia: An OpenTelemetry-Based Governance Platform for AI Systems

Introduces Traccia, an OpenTelemetry-based governance platform designed to ensure AI system compliance with regulations like the EU AI Act.

AI/ML arXiv cs.AI

CIPHER: A Decoupled Exploration-Selection Framework for Test-Time Scaling of Data Science Agents

Introduces CIPHER, a framework for data science agents that uses test-time scaling and decoupled exploration-selection to reduce cascading errors.

AI/ML arXiv cs.AI

Chat2Scenic: An Iterative RAG-Based Framework for Scenario Generation in Autonomous Driving

Presents Chat2Scenic, an iterative RAG-based framework for generating autonomous driving scenario scripts in Domain Specific Languages.

AI/ML arXiv cs.AI

A Comparative Analysis of Machine Learning Models for Long and Short-Term Forecasting of the Egyptian Stock Market: A Focus on EGX30

Compares various ML models (GRU, XGBoost, KNN) for predicting stock trends in the Egyptian EGX30 market.

AI/ML arXiv cs.AI

CatalogAgent: A Supervisor-mediated Self-Learning System Enabling Context Engineering for GenAI Models

Introduces CatalogAgent, a supervisor-mediated self-learning system that uses context engineering to improve e-commerce catalog enrichment.

Other Hacker News

How Has Roman Concrete Lasted for Millennia? 1,900-Year-Old Latrine Offers Clues

An exploration into the longevity of ancient Roman concrete, using a 1,900-year-old latrine as a case study for material science.

AI/ML arXiv cs.AI

Enhancing Small Language Models Reasoning through Knowledge Graph Grounding

Research on improving Small Language Models (SLMs) like Gemma 3 and Llama 3.2 using a neuro-symbolic agentic framework with knowledge graph grounding.

AI/ML arXiv cs.AI

Orchestrating Power Grid Studies with Multi-Agent AI and MCP Servers

A proposal for using Multi-Agent AI and the Model Context Protocol (MCP) to orchestrate power grid studies, featuring the pypowsybl-mcp interface.

AI/ML arXiv cs.AI

MemoHarness: Agent Harnesses That Learn from Experience

Introduction of MemoHarness, an adaptive framework that allows AI agent harnesses to learn and optimize from their own execution experience.