Deep InfoMax: The Bilinear Discriminator
Intermediate
The problem, from first principles
Deep InfoMax scores pairs of local and global features with a learned bilinear discriminator. The score is high when the local feature is consistent with the global summary, which is what the mutual information estimate rewards.
From theory to code
Implement discriminator_score(local, global_vec, W), returning local^T W global.
Constraints
Whas shape(d_l, d_g).
Hints
Hint 1
Multiply the local vector by the matrix, then dot with the global vector.
Keywords
research-papersunsupervisedmutual-informationrepresentation
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.