MIT Compbio Lecture 12 - Deep Learning (Fall '19)

MIT Computational Biology: Genomes, Networks, Evolution, Health

Prof. Manolis Kellis

Full playlist with all videos in order is here

All slides from Fall 2019 are here

Outline for this lecture:

1. Supervised Learning with Neural networks

- Perceptron, layers, activation units (sigmoid, softplus, ReLU)

- Learning: Gradient, Back-propagation, Rate, Dropout, Overfitting

2. Unsupervised learning with Deep belief networks & autoencoders

- Boltzmann machines, Restricted BMs (RBMs), Deep belief networks

- Learning: Energy, Gibbs Sampling, Simulated Annealing, Wake-sleep

3. Modern deep learning architectures

- Auto-encoders: Self-training, representation learning, RBM pre-training

- Convolutional neural networks: convolutional filters, pooling (sum/max)

- Recurrent neural networks: learning linear/temporal relationships

- Transfer learning, generative adversarial networks (GANs)

4. Deep learning in regulatory genomics

- Deciphering tissue-specific splicing code

- Deciphering regulatory grammars. DeepBind, DeepSea, Basset.

5. Three-dimensional structures: protein-DNA interactions, drug design

- Modern deep learning computing infrastructure

- Engines: TensorFlow, Theano, Torch, Caffe. Envs: Keras, Lasagene