Suite of multitask learning methods and a framework for easy experimentation with machine learning methods.

Title: Multi-Task Learning Software Suite (MTL4C), Version: 1.0

This repository contains a general purpose multitask learning (MTL) software suite implemented in Python. Classification and regression methods are included. In the MTL setup, a set of related tasks (regression or classification, for example) need to be solved and all models are learned jointly through a shared representation that allows for information transference accross related tasks. Notice that in MTL there is still a model associated with each task, the crucial point is the joint learning.

Two baseline methods are also implemented, namely: single task learning (STL), where a model is trained for each task independently; and pooled model, where a single model is trained for all tasks. The machine learning methods available in this software are located the folder named methods. Linear and non-linear models are available.

Two demo files are available in the experiments folder: demo_regression.py and demo_classification.py. One can build up on these files to run new experiments.

python demo_classification.py

• André R. Gonçalves (goncalves1@llnl.gov)
• David P. Widemann (widemann1@llnl.gov)
• Braden C. Soper (soper3@llnl.gov)
• Pryiadip Ray (ray34@llnl.gov)
• Jan Nygard (Jan.Nygard@kreftregisteret.no)
• Mari Nygard (Mari.Nygard@kreftregisteret.no)
• Ana Paula Sales (deoliveirasa1@llnl.gov)

Gonçalves, A.R., Von Zuben, F. J, and Banerjee, A. "Multi-task sparse structure learning with gaussian copula models." The Journal of Machine Learning Research 17.1 (2016): 1205-1234. paper

LLNL-CODE-773785