New Asynchronous Perception Machine enables efficient test-time-training

Researchers have introduced the Asynchronous Perception Machine (APM), a novel architecture designed for efficient test-time-training (TTT). APM can process image patches in any order, enabling it to recognize out-of-distribution images without prior dataset-specific training. The system demonstrates competitive performance against existing TTT methods by distilling a single representation from test samples to predict semantically-aware features. Beyond TTT, APM shows potential for semantic clustering of 2D images in a single pass and offers empirical support for the idea that input percepts function as fields. AI

IMPACT Introduces a novel architecture for efficient test-time-training, potentially improving out-of-distribution image recognition and semantic clustering capabilities.

RANK_REASON The cluster contains a research paper detailing a new machine learning architecture. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Asynchronous Perception Machine
• DagsHub
• Hugging Face
• Rajat Modi
• Test-Time Training on Graphs with Large Language Models (LLMs)

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →