Google DeepMind's WaveNet model: Wavelets and Learnable Filters

• Grant number: 25/21355-9
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Start date: December 15, 2025
• End date: March 01, 2026
• Field of knowledge: Physical Sciences and Mathematics - Probability and Statistics - Statistics
• Principal Investigator: Aluísio de Souza Pinheiro
• Grantee: Lívia Yumi de Souza Tuzita Yamamoto
• Supervisor: Abdourrahmane Atto
• Host Institution: Instituto de Matemática, Estatística e Computação Científica (IMECC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Institution abroad: Université Savoie Mont Blanc, France
• Associated to the scholarship: 25/08017-7 - Wavelet Clustering and Classification, BP.IC

Abstract

The idea for the project is to build a statistical procedure that combines the advantages of wavelets and Google DeepMind's WaveNet model through learnable filters. The main points which will be adressed by the student are: (a) general aspects of the WaveNet algortihm, (b)full study of its intermediate equations, (c) highlight the "wavelet/ multiscale" behavior of the aforementioned model, (d) identify the model's shortcomings. The student will then complement this model with wavelet filters that will compensate for the shortcomings by feeding back information in the form of specific multiscale wavelet coefficients to be identified, and train this new model in the sense of the optimal combination of WaveNet and wavelet contributions. An application to real data will be done as well.