AI/ML arXiv cs.AI

LakeQuest: A Three-Domain Benchmark for Grounded Question Answering across Data Lakes

LakeQuest is introduced as a benchmark for grounded question answering over heterogeneous, weakly structured data lakes.

AI/ML arXiv cs.AI

Evaluating Health Misinformation in Low-Resource Languages: Integrating Small Language Models with a Culturally-Sensitive Responsible NLP Framework (Bangla as a Case Study)

Research proposes a culturally-sensitive AI framework for detecting health misinformation in low-resource languages, using Bangla as a case study.

AI/ML arXiv cs.AI

Lost in Visual Translation: A VLM-Assisted Perceptual-Semantic Coherence Framework for EEG-to-Image Reconstruction

A new BCI-aware framework introduces the BCI-Coherence Score (BCS) to better evaluate EEG-to-image reconstructions by focusing on semantic recoverability.

Software Engineering Hacker News

Combinatorial Games in Lean

A discussion on applying the Lean theorem prover to combinatorial games.

Other Hacker News

America pays workers just 27% of what its wealth allows – the worst in the OECD

Analysis of wealth distribution and worker pay in America compared to other OECD countries.

AI/ML arXiv cs.AI

Self-Consistent Flow: Unifying Velocity and Endpoint Prediction for Rectified Flow Models

Introduces Self-Consistent Flow (SC-Flow), a method that unifies velocity and endpoint prediction to improve generative model training and sampling stability.

AI/ML arXiv cs.AI

From Reconstruction to Interpretation: Zero-Setup Multi-Phase Segmentation of X-ray Tomography Data

A zero-setup framework for multi-phase segmentation of X-ray tomography data that operates without user input or retraining.

AI/ML arXiv cs.AI

TRAIL: A Platform for Configurable Human--AI Teaming Experiments

Presents TRAIL, a web platform designed for configurable and reproducible human-AI teaming experiments.

AI/ML arXiv cs.AI

Comparing Semantic Navigation in Humans and Large Language Models using Natural Language Processing

A comparative study showing that human semantic search is more exploratory and variable than that of current LLMs.

AI/ML arXiv cs.AI

The Benjamini--Hochberg Procedure Can Fail to Control the FDR for Correlated Two-Sided Gaussian Tests

A theoretical proof disproving a 20-year-old conjecture regarding the Benjamini-Hochberg procedure's ability to control FDR for correlated Gaussian tests.

AI/ML arXiv cs.AI

RCWT: Measuring Task-Budget Displacement from Coordination Content in LLM Calls

Introduces the Roundtable Context Window Test (RCWT) to measure how coordination content displaces task budget in LLM prompts.

AI/ML arXiv cs.AI

Partial Identification with Multiple Nonlinear Measurements of a Latent Regressor

Research on partial identification of latent regressors using multiple nonlinear measurements, applied to AI occupational exposure.

AI/ML arXiv cs.AI

Fin-Analyst at FinMMEval 2026 Task 3: A Live Hybrid Trading Agent with LLM Specialists and Rule-Based Signals

Fin-Analyst, a hybrid trading agent combining LLM specialists and rule-based signals, achieved top ranking for TSLA in FinMMEval 2026.

AI/ML arXiv cs.AI

An Empirical Analysis of Continual Learning for Heterogeneous Medical Visual Question Answering

A systematic evaluation of continual learning (CL) methods for Medical Visual Question Answering (MedVQA) shows that existing methods struggle with stability-plasticity balance when tasks have diverse objectives.

AI/ML arXiv cs.AI

Representation and Reference Selection in Training-Free Synthetic Image Attribution

Research on synthetic image attribution suggests that attribution accuracy peaks at intermediate representation levels of models like CLIP and DINOv2, highlighting the importance of reference selection.

Cybersecurity arXiv cs.AI

AutoTrace: From Patches to Triggers via Agentic Interprocedural Exploration

AutoTrace is an agentic pipeline that localizes vulnerability triggers in code using a code property graph and LLM agents, outperforming previous state-of-the-art methods.

AI/ML arXiv cs.AI

Enabling 24-hour Agricultural Robotics: Unsupervised Day-to-Night Cross-Modal Image Translation for Nighttime Visual Navigation

AgriNight introduces an unsupervised image translation framework and a new benchmark dataset to enable autonomous agricultural robotics navigation during nighttime.

AI/ML arXiv cs.AI

Calibrated Selective Prediction Using Deep Ensembles for ROI-Based Thyroid Nodule Ultrasound Classification Under Dataset Shift: A Retrospective Evaluation

A retrospective evaluation of a deep ensemble framework for thyroid nodule classification shows strong internal calibration but limited external transportability under dataset shift.

AI/ML arXiv cs.AI

Sparse Autoencoders for Interpretable Out-of-Distribution Detection

This paper proposes using sparse autoencoders (SAEs) to learn interpretable features from intermediate network layers to improve out-of-distribution (OOD) detection.

AI/ML arXiv cs.AI

PFAdapter: Hierarchical LoRA Decomposition for Personalized Federated MLLMs

PFAdapter is a communication-efficient framework for personalized federated multimodal large language models that uses hierarchical LoRA decomposition to separate global and local parameters.