If they are still available, I would like to have a voucher to do some long term ML training. My dataset is getting quite huge, and my laptop's CPU is good enough for the actual playing, but for training, it's not cutting it anymore, and I would like to train it on some faster CPU (or GPU if available).
If anyone is interested, I'm using a variation on DQN and the Starcraft II shared knowledge/multi agent scaling whitepaper. Essentially, there is a ConvNet that represents some kind of general strategy/decision maker, and a NN that takes the ConvNet's output, and a ship's current state, and output a specific action for this ship. The goal is for the ConvNet and the NN to learn to talk to each other, so that effectively, there is a general higher-level strategy (the ConvNet), and a lower level/micromanagement with the Neural Network. There will be some shared experience with all of the ship's results/output feeding back into the ConvNet's thus hopeuflly accelerating the learning of the ConvNet talking to the NN, but so far, it has not been very successful. It has learned to (mostly) avoid other ship, but it has not learned about docking yet, and spends most of its time doing nothing.
My bottlenecks right now is time (Fulltime job + kids + laptop's CPU), and with the voucher, I feel that it would fix the CPU part of it : Faster CPU means generating data faster, and faster (hopefully way faster) training.