Reviews: TADAM: Task dependent adaptive metric for improved few-shot learning

I have read the rebuttal and will maintain my score. The submission and rebuttal motivate three orthogonal improvements (metric scaling, task-conditioning, auxiliary task co-training) to Prototypical Networks. The paper would be much better written as three distinct units that evaluate the three components separately and in more depth. The mathematical analysis section of the paper consists of writing out the gradient of the objective function with the additional temperature parameter. Although this provides intuition **about the limiting cases of \alpha**, I would hesitate to call this a significantly "non-trivial" derivation.