Toggle navigation sidebar

Toggle in-page Table of Contents

Few-shot and Zero-shot Learning for Music Information Retrieval

Few-Shot and Zero-Shot Learning for Music Information Retrieval

Introduction

What is Few-Shot Learning (FSL) and Zero-Shot Learning (ZSL)?
Advantages of FSL and ZSL in MIR

Foundations: Few-Shot Learning

Few-Shot Learning Foundations
Approaches
Metric-Based Few-Shot Learning
Optimization-Based Few-Shot Learning

Foundations: Zero-Shot Learning

Zero-Shot Learning Foundations
Zero-shot learning task formulation
Side Information

Few-Shot Learning in PyTorch

Introduction: Few-Shot Learning in PyTorch
Building a Class-Conditional Dataset
Sampling Few-Shot Learning Episodes
Building a Prototypical Network
Training a Few-Shot Instrument Classifier
Evaluating (and Visualizing) a Trained Prototypical Net

Zero-Shot Learning in PyTorch

Introduction: Zero-Shot Learning in PyTorch
Prepare dataset and splits
Models and data I/O
Word-audio ZSL training
Word-audio ZSL evaluation
Image-audio ZSL training
Image-audio ZSL evaluation

Recent Advances

Recent Advances in MIR
Classification: Musical Instrument Recognition
Classification: Sound Event Detection
Classification: Music Classification and Tagging
Drum Transcription
Music Source Separation

Conclusions

Future Directions
Summary

References

References

Powered by Jupyter Book

repository
open issue

Index

By Yu Wang, Hugo Flores García, and Jeong Choi
© Copyright 2022.