Transfer Learning Techniques

Transfer learning encompasses techniques that enable data scientists and engineers to build high-performing deep learning models without the massive computational and data labeling costs typically required when training from scratch.

This section examines two distinct subdomains of transfer learning, depending on the availability of labeled data and the degree of similarity between the source and target tasks.

Transductive transfer learning

This tutorial focuses on scenarios where labeled target data is available, the source and target tasks are the same, but the domain changes. For example, a sentiment analysis model trained on survey response data may need to be adapted to understand the nuances of product reviews.

Unsupervised transfer learning

This tutorial addresses the challenge of data scarcity. Unsupervised transfer learning techniques are helpful for adapting a model to a new domain when you have a target dataset, but no labels. For example, a model trained to detect anomalies in satellite images may need to be adapted to identify instances of illegal deforestation; however, the target dataset is large enough that hand-labeling the data is a significant challenge.