Learning Unsupervised Learning Rules

This repository contains code and weights for the learned update rule presented in "Learning Unsupervised Learning Rules." At this time, this code can not meta-train the update rule.

Structure

run_eval.py contains the main training loop. This constructs an op that runs one iteration of the learned update rule and assigns the results to variables. Additionally, it loads the weights from our pre-trained model.

The base model and the update rule architecture definition can be found in architectures/more_local_weight_update.py. For a complete description of the model, see our paper.

Dependencies

absl, tensorflow, sonnet

Usage

First, download the pre-trained optimizer model weights and extract it.

# move to the folder above this foldercd path_to/research/learning_unsupervised_learning/../# launch the eval scriptpython -m learning_unsupervised_learning.run_eval --train_log_dir="/tmp/learning_unsupervised_learning" --checkpoint_dir="/path/to/downloaded/model/tf_graph_data.ckpt"

Contact

Luke Metz, Niru Maheswaranathan, Github: @lukemetz, @nirum. Email: {lmetz, nirum}@google.com