Research/Blog
An Introduction to Meta Learning – a path to Artificial General Intelligence
- July 21, 2021
- Posted by: vsinghal
- Category: Artificial General Intelligence (AGI) papers
Author : Bhavesh Laddagiri
Submitted on : 5 Sep 2020
Abstract – Meta Learning (aka Learning to Learn) is about designing models which can learn and adapt to new tasks fast and efficiently without much fine-tuning. We want to design models that can learn to generalize well to new tasks, data or environments which are different from the ones in training. E.g. :
1. A game playing bot trained to play PUBG should be able to generalize to Call of Duty
2. A robot trained to walk on flat soil should be able to walk on ragged terrain
3. An image classifier trained on dogs should be able to recognize cats given a few images of what a cat looks like
Types of Meta-Learning
Meta Learning can be approached in different ways :
1. Metric-Based – Learn an efficient distance function for similarity
2. Model-Based – Learn to utilize internal/external memory for adapting (MANN)
3. Optimization-Based – Optimize the model parameters explicitly for learning quickly
Keywords – Meta Learning, Artificial General Intelligence, Metric-based Meta Learning, Model-based Meta Learning, Optimization-based Meta Learning