ConditionalMetaLearning

We propose a conditional Meta-Learning approach to Biased Regularization and Fine Tuning for heterogeneous tasks.

This repository is the official implementation of the paper:

'The Advantage of Conditional Meta-Learning for Biased Regularization and Fine Tuning' (ID: 6479)

The scripts are organized in the following folders/files.

1. 'data' folder: it contains the Schools and the Lenk dataset we used for our real experiments.
2. 'saved_results': it contains the results we got.
3. 'src' folder: it contains the following files. A) 'data_management.py': it generates the data for the different experimental settings. B) 'general_functions.py': it contains the basic functions used (such as loss, loss subgradient, feature map). C) 'inner_algorithm.py': it contains the implementation of the online inner algorithm (fine tuning variant). D) 'methods.py': it contains the implementation of the inner algorithm with a fixed meta-parameter vector (constant conditioning function), the implementation of the unconditional meta-learning approach and the implementation of the conditional meta-learning approach.
4. 'post_processing.py': it allows to plot the results memorized in 'saved_results' folder.
5. 'main_script.py': it allows to run the methods on the different experimental setting.