Reproduce “Memory Efficient Meta-Learning with Large Images” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
John Bronskill, Daniela Massiceti, Massimiliano Patacchiola, Katja Hofmann, Sebastian Nowozin, Richard E. Turner
LITE enables meta-learners to train on large support sets by processing all examples for task adaptation while back-propagating query loss through only a random subset. On ORBIT, Simple CNAPS with EfficientNet-B0 achieves 82.7% frame accuracy on clean videos and 65.6% on clutter videos. It adapts in one forward pass and 2.92 seconds per task; clutter performance remains competitive.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.