Graduation project AutoML for time-series data

65 views

Skip to first unread message

Mitra Baratchi

unread,

Sep 24, 2021, 9:43:28 AM9/24/21

to LIACS thesis projects

Dear students,

Please find below a thesis research topic with the goal of improving AutoML for time-series data.

Regards,

Mitra

____________

Title: Meta-learning for wearable/time-series data facilitated by OpenML

Machine learning research can to a great extent benefit from meta-data and empirical machine learning experiment results on similar datasets. This meta-data can for instance be used for meta-learning and warm-starting automated machine learning models. Automated machine learning has benefited from such techniques when dealing with Tabular data. There is, however, limited research performed on expanding the use of these techniques for other formats of data such as time-series. Time-series data are ubiquitous and are becoming more relevant as it is possible to repeatedly collect data using sensing technologies in different domains (medical, environmental, etc.).

Our main goal in this project is to improve the state of automated machine learning models for wearable and time-series data using meta-learning on time-series datasets. To do so we aim to make use of open science platforms such as OpenML. This would allow us not only to facilitate collection of similar datasets, but also to gather knowledge on the types of tasks and the process of training models on such data.

Datasets/sensors available for this research:

Public benchmarking datasets

Steps of the research project:

Collecting a relevant benchmark suite of time-series data
Identifying meta-learning techniques for time-series data and incorporating them in available AutoML systems
Extending the OpenML system to facilitate collection and time-series data and performing machine learning tasks on them