Question about Machine learning on Online Neuroscience data

I'm doing a BSc in Artificial Intelligence and for my final year project I was considering doing "Predicting the onset of epileptic seizures". To do this I'm meant to use DSP and/or AI techniques to process "episodic activity" captured from electrodes implanted into the brains of a few suffers of epilepsy. However my main AI knowledge hasn't really gone past Neural Networks (multilayer back prop), Radial basis functions, self organising maps until now .

Now I'm interested in using deep learning and I was initially thinking of using something like an autoencoder to reduce the dimensionality and hopefully pretraining on some known data. Though aside from that in my mind I’m still running the compressed data through a standard Multilayer perceptron.

I was also thinking of using Simulated Annealing as opposed to back propagation for training the encoder and whatever layers follow it. I’ve read interesting stuff about it and seen promising data and the maths should be fairly simple to implement.

I was also debating between C++ and python for implementation, C++ I’m very comfortable with but I started learning python a few months back and I like the conciseness.

Any help, advice or criticism of my initial method would be welcome.