Graph-based and transfer learning models for times series representation

• Grant number: 23/08087-0
• Support Opportunities: Scholarships abroad - Research Internship - Master's degree
• Start date: October 01, 2023
• End date: December 17, 2023
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: Daniel Carlos Guimarães Pedronette
• Grantee: Bionda Rozin
• Supervisor: Ricardo da Silva Torres
• Host Institution: Instituto de Geociências e Ciências Exatas (IGCE). Universidade Estadual Paulista (UNESP). Campus de Rio Claro. Rio Claro , SP, Brazil
• Institution abroad: Wageningen University & Research, Netherlands
• Associated to the scholarship: 22/01359-1 - Times series retrieval through unsupervised learning, BP.MS

Abstract

With the wide applicability of time series data in various domains, including medicine, agriculture, economics, and science, the analysis and processing of such data have become increasingly demanding. Information Retrieval tasks applied to time series data play a crucial role in identifying patterns and ranking data based on similarity, making them a key focus in the area. Machine Learning tasks, such as information retrieval, classification, and clustering, heavily rely on effective computational representations of data to generate more accurate results and well-founded conclusions. Hence, one of the challenges in the area is obtaining efficient representations for Time Series data. In this project, our primary objective is to explore novel approaches for time series representation. We aim to model problems using graph-based techniques and derive a multivariate time series representation by decomposing each dimension and treating it as a univariate time series, followed by a rank aggregation task, which combines each representation into a unified one. Furthermore, we intend to develop a feature extractor for time series using deep learning methods, based on the advancements made in image feature extraction utilizing Convolutional Neural Networks (CNNs). This task is expected to be the most challenging aspect of our project. By addressing these objectives, we aim to advance the field of time series analysis and contribute to the development of more robust and effective techniques for machine learning tasks in the area. (AU)