Abstract: Constructing a good machine learning model is complicated and time-intensive. Relevant data has to be collected and cleaned, features might need to be engineered, the right machine learning algorithms have to be chosen, their hyperparameters have to be tuned, and the resulting model’s performance should be evaluated. Automated Machine Learning (AutoML) aims to automate these steps. In our workshop we will focus on the state of the art in automatic machine learning pipeline construction and neural architecture search (NAS), and provide hands-on guidance on using different methods that are used to optimize pipelines, such evolutionary optimization and Bayesian optimization. Much like human machine learning experts learn from experience, we also want to leverage prior knowledge to find better models, faster. This is also known as meta-learning, which we apply in an AutoML context by learning which hyperparameters are more important, or which pipeline structures or neural architectures should be explored.
Module 1: State of the art in AutoML. A mini-tutorial on today's leading AutoML approaches and a more in-depth look on key approaches and available tools.
Module 2: Learning from machine learning experiments (hands-on). We show how to use `openml-python` to download datasets and available machine learning experiment results from OpenML.org. We visualize the effect of hyperparameters without running any experiments but also show how to create and share your own experiments with OpenML.
Module 3: AutoML in Python. To help us get a better overview of the available AutoML systems, we introduce you to our AutoML Benchmark which evaluates various systems across a wide range of datasets. After this session you will be able to build a model with AutoML in Python, and know how to get insight on how the different tools work for your data.
General understanding of the most important machine learning algorithms. Good coding experience in Python.
Bio: Pieter Gijsbers is a PhD student at the Eindhoven University of Technology. His areas of interest are automated machine learning and meta-learning. He is the main author of GAMA, a research-focused open source AutoML tool. While GAMA is easy to use for end-users, it also allows researchers to try different search techniques and visualize the optimization process. He is a co-author of the Open Source AutoML Benchmark. He is also part of the team working on the openml-python package, providing an easy to use Python interface to OpenML.