Hands-On One-shot Learning with Python

Optimization-Based Methods

Most deep learning models learn objectives using the gradient-descent method; however, gradient-descent optimization requires a large number of training samples for a model to converge, which makes it unfit for few-shot learning. In generic deep learning models, we train our models to learn to accomplish a definite objective, whereas humans train to learn any objective. Following this observation, various researchers have created different optimization approaches that focus on learn-to-learn mechanisms.

In other words, the system focuses on how to converge any loss function (objective) instead of minimizing a single loss function (objective), which makes this algorithmic approach task and domain invariant. For example, ...