Clarification of term episodic vs meta-set vs meta-batching

I think I asked this somewhere but can’t find it easily and I think it’s important.

When we have a meta-set (a data-set of data-sets with support & query splits) D_meta = {D_i}^N_{i=1} is the index i called an episode or is an episode a (meta) batch (of data-sets) from this meta-set (e.g. of size 16)?

cross posted: https://stats.stackexchange.com/questions/478255/what-does-the-term-episode-mean-in-meta-learning

reference:

• https://github.com/tristandeleu/pytorch-meta/issues/78
• https://www.quora.com/unanswered/What-does-the-term-episode-mean-in-meta-learning
• https://www.reddit.com/r/MLQuestions/comments/hve478/what_does_the_term_episode_mean_in_metalearning/?
• https://discuss.pytorch.org/t/what-does-the-term-episode-mean-in-meta-learning/90051
• https://stats.stackexchange.com/questions/478255/what-does-the-term-episode-mean-in-meta-learning

[tristandeleu]

Usually an episode means one single dataset D_i. If you have a (meta-)batch of size 16, this means you have 16 episodes/datasets in your (meta-)batch.

[tristandeleu]

I agree with your definition of batch of tasks, but this does not correspond to an episode (see my comment above https://github.com/tristandeleu/pytorch-meta/issues/78#issuecomment-662343834). Do you have a reference for an episode being a batch of tasks?

Defining an episode as being a single task refers to episodic learning, where your episode represents the whole experience you get to see to learn to solve a task.