Create a 'masking' blob for new loss layer?

[Nicholas Dufour]

Hi All-

Two quick questions:

- What is the best way to instantiate a new blob Y for a loss layer, that will be persistent, that will be a square N x N matrix?
- Also instantiate a constant blob Z, representing another N x N matrix, that will serve as a 'masking' blob for Y, such that we modify Y to yield Y' where Y'[i, j] = Y[i, j] iff Z[i, j] and 0 otherwise.

Thanks!