Software

This page provides an overview of the software that we built with the goal of transforming the way people do (automated) machine learning research in a way that is more open, reproducible, and accessible. Most of the software below is developed with collaborators outside of our lab.

We also open-source the code for our papers under our GitHub organisation. Code for each paper is also linked in the overview in the papers page.

OpenML

Machine learning research should be easily accessible and reusable. OpenML is an open platform for sharing datasets, algorithms, and experiments - to learn how to learn better, together.

Website Github

AutoML Benchmark

The AutoML Benchmark (AMLB) is software that performs end-to-end evaluation of AutoML frameworks. It can be used to compare current state-of-the-art and evaluate new methods with ablation studies. Over 10 AutoML frameworks can be evaluated out-of-the-box on over 100 datasets, and it’s easy to bring your own data or AutoML framework.

Website Github

GAMA

The General Automated Machine learning Assistant (GAMA) was developed to be able to easily experiment with new AutoML design through ablation studies by providing a modular AutoML framework. We have sunset the project, and decided to focus our efforts on OpenML and AMLB. If you are interested in a modern implementations driven by similar philosophies, please have a look at the AutoML Toolkit.

Website Github

SERENA

Self-Regulated Neurogenesis for Online Data-Incremental Learning (SERENA) is a neuro-inspired lightweight and efficient method for Online Data-Incremental Learning, designed to continually adapt to streaming data without forgetting the past knowledge.

Github

CL-with-DST

Continual Learning with Dynamic Sparse Training (CL-with-DST) investigates the evolution of sparse network topologies within the continual learning framework, where data arrives sequentially in a streaming fashion. This approach dynamically adapts the sparse structure of the network over time, enabling efficient learning from non-stationary data while mitigating forgetting.

Github