Code accompanying the ICML-2018 paper "Gradient-Based Meta-Learning with Learned Layerwise Metric and Subspace"
Switch branches/tags
Nothing to show
Clone or download
Latest commit fe3f8c7 Oct 6, 2018
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data add data handling scripts Jun 16, 2018
experiments Rename omni20.sh to omniglot.sh Jun 16, 2018
LICENSE add code Jun 16, 2018
README.md Update README.md Oct 7, 2018
__init__.py add code Jun 16, 2018
data_generator.py add code Jun 16, 2018
main.py add code Jun 16, 2018
maml.py add code Jun 16, 2018
poly_generator.py add code Jun 16, 2018
special_grads.py add code Jun 16, 2018
utils.py add code Jun 16, 2018

README.md

MT-net

Code accompanying the paper Gradient-Based Meta-Learning with Learned Layerwise Metric and Subspace (Yoonho Lee and Seungjin Choi, ICML 2018). It includes code for running the experiments in the paper (few-shot sine wave regression, Omniglot and miniImagenet few-shot classification).

Data

For the Omniglot and MiniImagenet data, see the usage instructions in data/omniglot_resized/resize_images.py and data/miniImagenet/proc_images.py respectively.

Usage

To run the code, see the usage instructions at the top of main.py.

For MT-nets, set use_T, use_M, share_M to True.

For T-nets, set use_T to True and use_M to False.


This codebase is based on the MAML repository.