Experimenting deep learning-based strategies to deal with multivariate time series tasks.

• Grant number: 23/11745-9
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Effective date (Start): December 01, 2023
• Effective date (End): February 29, 2024
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Diego Furtado Silva
• Grantee: Andre Guarnier De Mitri
• Supervisor: Germain Forestier
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Research place: Université de Haute-Alsace, France
• Associated to the scholarship: 23/05041-9 - Adapting Time Series Classification Algorithms to Regression, BP.IC

Abstract

With the growing ubiquity of smartphones, smartwatches, and other devices capable of collecting data across time, applications that use time series as input, such as cardiac monitoring and activity recognition, have become increasingly popular. Several of these scenarios can be mapped naturally as tasks involving multivariate time series. Thus, various techniques for Machine Learning applied to time series have been developed. However, most of the developed techniques for multivariate time series were adapted from univariate data. However, there is no clear guidelines to choose the best adaptation for deep learning multivariate time series classification and extrinsic regression due to the lack of standardize experimental evaluation. Considering this scenario, this research will propose, implement, and execute an experimental evaluation procedure to assess different techniques to adapt deep neural networks designed for univariate time series classification and regression to deal with multivariate problems. (AU)