Reviews: DeepMath - Deep Sequence Models for Premise Selection

Pros: * The problem is interesting and does not appear to have previously received attention from the NIPS community * The paper is clearly written Cons: * The approach is not particularly novel; it is mostly a case of applying a range of existing neural network components to the premise selection problem. Is there a way to meaningfully compare these numbers? If not, it's confusing to include them. This presentation is asking us to take a maximum of test performance over all the different neural network variants, and to compare that to the performance of a single baseline method. However, there is quite a lot of variance in the neural network performance relative to the difference between neural network and baseline performance.