MLRF Machine Learning Research Flashcards

Multiple Choice Learning: Learning to Produce Multiple Structured Outputs

Read Paper

What is a common technique used to calculate the loss for models with multiple outputs (e.g. Segment Anything)?

Compute the loss for each of the predicted outputs, but only backpropagate the lowest loss.

Let Y^i={y^i1,...,y^iM}\hat{Y}_i = \{\hat{y}^1_i, ..., \hat{y}^M_i\} be the set of predicted outputs for input xix_i. L(Y^i)=min⁡y^i∈Y^il(yi,y^i)\mathcal{L}(\hat{Y}_i) = \operatorname{min}_{\hat{y}_i \in\hat{Y}_i} \mathcal{l}(y_i, \hat{y}_i)

Machine Learning Research Flashcards is a collection of flashcards associated with scientific research papers in the field of machine learning. Best used with Anki or Obsidian. Edit MLRF on GitHub.