research

Continual Learning

As machines continue to advance and take on more complex tasks, their ability to learn and adapt over time becomes increasingly crucial. This is where my area of interest and expertise, continual learning, comes into play. Continual learning refers to a machine’s ability to acquire new skills and knowledge without forgetting what it previously learned. This poses specific challenges in the field of deep learning, as neural networks often have a tendency to “forget” when trained with new information.

Meta Learning

Meta-learning deals with a collection of tasks. The basic idea is to learn about how to learn novel tasks efficiently. Given a distribution of tasks, typically an agent or learner is trained to capture the commonality among these tasks. This allows the learner to quickly adapt to new, unseen tasks with similar structure using only a few samples or few update steps.

Continual Learning \(\cap\) Meta-Learning

The intersection of continual and meta-learning introduces intriguing research avenues. The main idea is to use meta-learning mechanisms or anything related to the learning process that are “good” for learning from a stream of experiences via meta-objectives, without explicitly defining the rules.