Using fastai to classify Japanese kanji characters

How to deal with imbalanced data

Introduction

This article describes how to use fastai for multiclass classification, specifically 3,017 classes of Japanese kanji characters. Kanji characters are a major part of the Japanese writing system alongside hiragana and katakana. There are many thousands of kanji characters, which makes it a challenging way to explore multiclass classification models. All the experiments have been executed in Google Colab. I included various code snippets in this article, the …

fastai imbalanced-data japanese transfer learning