AI/ML arXiv cs.AI

Large-Language-Model Discovery of Quantum LDPC Codes through Structured Concept Evolution

Uses LLMs and a structured mutation grammar (SCE) to discover new quantum low-density parity-check (qLDPC) codes.

AI/ML arXiv cs.AI

IV-CoT: Implicit Visual Chain-of-Thought for Structure-Aware Text-to-Image Generation

Proposes IV-CoT, a latent visual reasoning framework to improve structure-aware text-to-image generation using implicit visual chain-of-thought.

AI/ML arXiv cs.AI

It's Complicated: On the Design and Evaluation of AI-Powered AAC Interfaces

Explores the design and evaluation of AI-powered interfaces for augmentative and alternative communication (AAC).

AI/ML arXiv cs.AI

FLUX3D: High-Fidelity 3D Gaussian Generation with Diffusion-Aligned Sparse Representation

Presents FLUX3D, a scalable image-to-3DGS framework improving 3D Gaussian Splatting reconstruction fidelity and alignment.

Other Hacker News

Exploring the internal representations of Pangram 3.3.2

An exploration into the internal representations of Pangram 3.3.2.

Other Hacker News

Ending All Respiratory Infections

A discussion regarding the effort to end all respiratory infections.

AI/ML Hacker News

Bible as RAG Database

An implementation or discussion of using the Bible as a Retrieval-Augmented Generation (RAG) database.

Software Engineering Hacker News

Mixing Visual and Textual Code

Research or a proposal on mixing visual and textual representations of code.

Other Hacker News

MSc Thesis – The Limits of Generalized Sync

An MSc thesis exploring the theoretical limits of generalized synchronization.

AI/ML arXiv cs.AI

Task Decomposition for Efficient Annotation

Proposes a method to decompose complex structured annotation tasks into sub-tasks to reduce inferential load and improve cost-efficiency.

AI/ML arXiv cs.AI

Beyond U-Net: A Latent-Representation-Aligned Skip-Free Backbone for Flow-Matching Speech Enhancement

Introduces a skip-free encoder-decoder backbone for flow-matching speech enhancement to improve real-time deployment and audio quality.

AI/ML arXiv cs.AI

UniDrive: A Unified Vision-Language and Grounding Framework for Interpretable Risk Understanding in Autonomous Driving

Presents UniDrive, a framework for autonomous driving that combines temporal reasoning and high-resolution perception for interpretable risk understanding.

AI/ML arXiv cs.AI

Context-Aware Prediction of Student Quiz Performance with Multimodal Textbook Features

Study on using multimodal textbook features (text and images) to improve the prediction of student quiz performance.

AI/ML arXiv cs.AI

DeepBD: A Grounded Agentic Workflow for Variant Prioritization and Diagnosis of Genetic Birth Defects

Introduces DeepBD, an agentic workflow for diagnosing genetic birth defects by prioritizing variants using LLMs and evidence engines.

AI/ML arXiv cs.AI

A Fair Evaluation of Graph Foundation Models for Node Property Prediction

Researchers reevaluated 9 Graph Foundation Models for node property prediction, finding that only the most recent Prior-data Fitted Networks outperform well-tuned Graph Neural Networks.

AI/ML arXiv cs.AI

Poster: Exploring the Limits of Audio-Based Detection of Turkish Phone Call Scams

A study on detecting Turkish phone scams using LLMs introduces a new multi-modal dataset and finds that transcript-based inputs outperform raw audio processing.

AI/ML arXiv cs.AI

Toward Self-Evolution-Ready Workflow Harnesses: A Reversible Migration Path and Convertibility Taxonomy for Expert LLM Pipelines

The paper proposes a reversible migration path using a Strangler-Fig approach and a convertibility taxonomy to transform legacy LLM workflows into composable, typed stages.

AI/ML arXiv cs.AI

Infinitesimal Causality

This theoretical paper introduces a categorical account of infinitesimal causality using Frobenius Markov categories and tangent-bundle semantics to model causal interventions.

AI/ML arXiv cs.AI

Privacy-Preserving RAG via Multi-Agent Semantic Rewriting: Achieving Confidentiality Without Compromising Contextual Fidelity

A new multi-agent framework for Privacy-Preserving RAG uses semantic rewriting to remove sensitive identifiers from retrieved content without losing contextual fidelity.

AI/ML arXiv cs.AI

Visualizing "We the People": Bridging the Perception Gap through Pluralistic Data Storytelling

This research explores using AI to create pluralistic data visualizations that emphasize nuance and consensus in public opinion, reducing political polarization.