AI/ML Hacker News

Show HN: BlitzGraph – Supabase for graphs, built for LLM agents

BlitzGraph is introduced as a graph database alternative to Supabase specifically designed for LLM agents.

AI/ML arXiv cs.AI

Temporal Preference Optimization for Unsupervised Retrieval

TPOUR is a new unsupervised retrieval method that uses Temporal Preference Optimization to better handle temporal relevance in document retrieval.

AI/ML arXiv cs.AI

FacProcessTwin: An LLM-Based System for Process Twin Development

FacProcessTwin leverages LLMs to automate the creation of process twins for manufacturing, reducing development time significantly.

AI/ML arXiv cs.AI

Handling Feature Heterogeneity with Learnable Graph Patches

This research proposes learnable graph patches to address feature heterogeneity in graph foundation models, improving transferability across different datasets.

AI/ML arXiv cs.AI

ASTEROID: A Spatiotemporal Information Transformer for Forecasting Multi-Step Time Series of Molecular Dynamics

ASTEROID is a spatiotemporal information transformer designed to predict multi-step atomic coordinates in molecular dynamics simulations, reducing computational costs.

AI/ML arXiv cs.AI

See First, Answer Later: Visual Evidence Pre-Alignment via Sufficiency-Driven RL

Visual Evidence Pre-Alignment (VEPA) is a new training stage for MLLMs to improve visual grounding using sufficiency-driven RL via GRPO.

AI/ML arXiv cs.AI

SuCo: Sufficiency-guided Continuous Adaptive Reasoning

SuCo is a training framework that implements continuous adaptive reasoning to reduce redundant Chain-of-Thought tokens in Large Reasoning Models.

AI/ML arXiv cs.AI

SegTME-UNI2: A Foundation Model-Based Framework for Generalisable Multiclass Cell Segmentation and LLM-Driven Tumour Microenvironment Characterisation in Histopathology

SegTME-UNI2 is a foundation model-based framework for multiclass cell segmentation and tumor microenvironment characterization in histopathology.

AI/ML arXiv cs.AI

Confusion-Aware Transfer Teacher Curriculum Learning Framework: Disentangling Scoring and Pacing Effects

Research on the Transfer Teacher framework explores confusion-aware difficulty scoring in curriculum learning, finding benefits primarily in data-efficient regimes.

AI/ML Hacker News

Midjourney Medical

Discussion regarding Midjourney's application in the medical field.

AI/ML Hacker News

Midjourney Ultrasonic CT Scanner

Exploration of using Midjourney for ultrasonic CT scanning visualization or conceptualization.

Other Hacker News

How Madrid built its metro cheaply

A look at the cost-effective engineering strategies used to build the Madrid metro system.

Tech Business/VC Hacker News

Brad Feld – Does the Rule of 40 Work for Hardware?

Brad Feld discusses whether the 'Rule of 40' financial metric is applicable to hardware companies.

AI/ML arXiv cs.AI

SkillMoV: Mixture-of-View Routing with Prototype-Conditioned Gating for Unified Multi-View Proficiency Estimation

Introduces SkillMoV, a parameter-efficient framework for multi-view proficiency estimation from synchronized video using a Mixture-of-View Projector.

AI/ML arXiv cs.AI

Divide, Deliberate, Decide: A Multi-Agent Framework for Fine-Grained Egocentric Action Recognition

Proposes a zero-shot multi-agent framework utilizing heterogeneous VLM specialists and a Borda count for fine-grained egocentric action recognition.

AI/ML arXiv cs.AI

Bounding Box Label Propagation for Re-Annotation of Document Layout Analysis Datasets

Presents Bounding Box Label Propagation (BBLP), a pseudo-labelling framework to reduce manual re-annotation efforts in document layout analysis.

AI/ML arXiv cs.AI

SketchXplain: Intuitive Visual Explanations of Image Classifiers with Sketches

Introduces SketchXplain, a system that generates intuitive sketch-based visual explanations for image classifiers to improve XAI interpretability.

AI/ML arXiv cs.AI

A Risk Decomposition Framework for Pre-Hoc Fine-Tuning Prediction

Develops a risk decomposition framework to theoretically predict the performance of LLM fine-tuning before it occurs.

AI/ML arXiv cs.AI

TuneAhead: Predicting Fine-tuning Performance Before Full Training Begins

Presents TuneAhead, a lightweight framework that predicts fine-tuning performance for LLMs using meta-feature vectors and SHAP-based diagnostics.

Other TechCrunch

How to turn off AI in your Google Docs

A guide on how to disable Gemini AI pop-ups and features within Google Docs.