Implement a simple machine learning experiment

Your job will be to set up an experiment for a simple (but novel) idea for a neural network application. The first stage will be to set up some configuration for fast training of a large data set. E.g. one of those here: [login to view URL] You'll choose one of these and set it up to run on a single local machine using its GPU. Next, you will implement an idea for further accelerating the training. To give a very rough analogy of what we're trying to achieve, take a look at [login to view URL] (our idea is completely different and will be shared with you under NDA.) Your code should be within a jupyter notebook running locally, using either a library (e.g. pytorch) that we'll agree on, and display key metrics (such as the learning curve) while the network is being trained. Although the training above implies using mostly convolutional nets, the work involves an RNN as well. When you submit, please concisely describe the most sophisticated RNN that you trained. What problem was it trying to solve, what was the network architecture you chose, how much training data did you use, etc