Meta-Learning of Structured Task Distributions in Humans and Machines
international conference on learning representations, 2021.
We developed a novel meta-learning task with a structured task distribution and statistically equivalent "null" task distribution to show humans are more adept at the former whereas current meta-learning agents are more adept at the latter.
In recent years, meta-learning, in which a model is trained on a family of tasks (i.e. a task distribution), has emerged as an approach to training neural networks to perform tasks that were previously assumed to require structured representations, making strides toward closing the gap between humans and machines. However, we argue that e...More
PPT (Upload PPT)