Continual Learning for Music Classification

Music classification is a prominent research area within Music Information Retrieval. While Deep Learning methods can adequately perform this task, their classification space remains fixed once trained, which conflicts with the dynamic nature of the ever-evolving music landscape. This work explores, for the first time, the application of Continual Learning (CL) in the context of music classification. Specifically, we thoroughly evaluate five state-of-the-art CL approaches across four different music classification tasks. Additionally, we showcase that a foundation model might be the key to CL in music classification. To that end, we study a new approach called Pre-trained Class Centers, which leverages pre-trained features to create fixed class-center spaces. Our results reveal that existing CL methods struggle when applied to music classification tasks, whereas this simple method consistently out-performs them. This highlights the need for CL methods tailored specifically for music classification.