Name: Transfer Learning Lab
Author: MySimulator

Setup

Labelled training examples: 10 Source-target task similarity: 0.8

A pretrained model already encodes useful features from a large source task. Fine-tuning adapts those features to a new target task using only a small labelled dataset, rather than learning everything from random weights.

Test accuracy after training

Fine-tuned (transfer)—

From scratch—

Test accuracy vs. training examples used