Future Directions#

In this section, we would like to highlight some remaining challenges and potential future directions for FSL and ZSL in MIR.

  1. Train FSL/ZSL models with less labeled data - Unsupervised FSL [46], semi-supervised FSL [47]

  2. Leverage existing pre-trained models - Few-shot continual learning [48][49]

  3. New FSL/ZSL methods

  4. User study - Support set selection process, interaction flow

  5. Apply FSL/ZSL to more MIR tasks - Recommendation, generation, problems that you are working on?