Task-Informed Meta-Learning

This repository contains examples of Task-Informed Meta-Learning (paper).

We consider two tasks:

• Crop Type Classification
• Yield Estimation
• Grouped Omniglot

Each task acts as its own self-contained codebase - for more details on running the experiments, please check their respective READMEs.

Getting started

For both tasks, anaconda running python 3.6 is used as the package manager. To get set up with an environment, install Anaconda from the link above, and (from either of the directories) run

conda env create -f environment.yml

Once the environment is activated, the main script to train the models is then deep_learning.py, with the model configurations controlled by the config.py file.

The trained TIML models are available on Zenodo.