This code implements and collections various contrastive losses for self-supervised learning. This code is under active development. An overview of the included:
Deep Graph Infomax [1] attempts to learn a bilinear
layer capable of discriminating between positive examples (any input
GraphTensor) and negative examples (the input GraphTensor but with perturbed
features: this implementation, as in the original paper, shuffles features
across batch, that is, the components the merged GraphTensor).
Deep Graph Infomax is particularly useful in unsupervised tasks that wish to learn latent representations informed primarily by a node's neighborhood attributes (vs. its structure).
- [1] Petar Veličković, William Fedus, William L. Hamilton, Pietro Liò, Yoshua Bengio, R Devon Hjelm: "Deep Graph Infomax", 2018.
TensorFlow programs can import and use this model as described in its API docs.
The API of this model may change between OSS library versions.