What do we mean by mutual exclusivity of tasks?

This work (E Pan, 21) and this one (M Yin, 20) state that most classification meta-learning algorithms fail for non-mutually exclusive tasks as the model may over-fit to a task, and no model can solve all the tasks at once (respectively).

I had trouble understanding the exact meaning of a "task" in meta classification here. [E Pan, 21] uses "task" synonymously with "new class", while [M Yin, 20] states "...prior work uses a per-task random assignment of image classes to N-way classification labels". However, some priors on few-shot learning [S. Hugo, 17], and [Y Wang, 19] agree with FFLab's, (20) description of "task" which I found more clear:

The number of classes (N) in the support set defines a task as an N-class classification task or N-way task, and the number of labeled examples in each class (k) corresponds to k-shot, making it an N-way, k-shot learning problem.

Where the support set $D_s$ here is part of the meta training data $D$ which comprises a support and test set $D_t$ $D = <D_s, D_t>$ [Weng, 18].

However, even with a better understanding of what a "task" is, I still couldn't get what constitutes mutually exclusive tasks.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.