Add Barlow Twins loss for representation learning

[Lucas-rbnt]

Description

Addition of the BarlowTwinsLoss class. This cost function is introduced in the http://proceedings.mlr.press/v139/zbontar21a/zbontar21a.pdf paper with the aim of disentangling the representations learned on two views of the same sample, making it a powerful tool for multimodal and unsupervised learning. This cost function is similar to the InfoNCE Loss function already implemented in MONAI (https://docs.monai.io/en/latest/_modules/monai/losses/contrastive.html#ContrastiveLoss). However, it differs in several respects: there is no l2-normalisation, but rather a z-normalisation. In addition, rather than working between pairs of embeddings, Barlow Twins seeks to decorrelate the components of the representations.

\mathcal{L}_{BT} := \sum_i (1 - \mathcal{C}_{ii})^2 + \lambda \sum_i \sum_{i\neq j} \mathcal{C}_{ij}^2

with $\lambda$ a positive hyperparameters and $\mathcal{C}$ the cross-correlation matrix

Types of changes

• [x] Non-breaking change (fix or new feature that would not break existing functionality).
• [ ] Breaking change (fix or new feature that would cause existing functionality to change).
• [x] New tests added to cover the changes.
• [x] Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.
• [x] Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.
• [x] In-line docstrings updated.
• [x] Documentation updated, tested make html command in the docs/ folder.