Reviews: Recurrently Controlled Recurrent Networks

I'm glad to hear that you are going to open source the optimizations; I look forward to playing with this. Best of luck with the follow-up work, and I look forward to seeing how the RCRN performs on SNLI in the cross-attention task setting (really hoping to see this in the camera-ready!). Original Review: The core idea behind this paper is to use RNNs (LSTMs or GRUs in this work) to compute the gates (input, forget, etc.) for a higher level RNN. They use this idea to show how to improve performance on a large number of tasks (mostly classification) with relatively simple models that have little to no loss in efficiency compared to models that perform similarly. Some of the performances achieved new state-of-the-art results.