Novel Architectures for Time Series Classification and Extrinsic Regression

• Grant number: 25/07753-1
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: September 01, 2025
• End date: August 31, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Diego Furtado Silva
• Grantee: Gabriel da Costa Merlin
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

The advancement of technologies such as the Internet of Things, smart cities, and Industry 4.0 has driven the generation of large volumes of time series data through sensors applied in various domains, such as healthcare and industry. In this context, extrinsic classification and regression tasks are becoming increasingly relevant for extracting knowledge from this data. This project proposes to investigate the use of recent deep learning architectures - especially Kolmogorov-Arnold Networks (KAN), based on functional representations, and Mamba, grounded in selective state space models - applied to these tasks. Adaptations of these architectures will be developed and evaluated, focusing on their applicability and performance compared to established models such as convolutional neural networks and Transformers. The research will utilize benchmark repositories (UCR and TSER) for experimental validation and aims to contribute to the advancement of robust and efficient techniques for time series analysis. The project also includes a planned research internship abroad (BEPE), in collaboration with researcher Tony Bagnall (University of Southampton), a specialist in the field. (AU)