Content:
– Motivational examples;
– Why use deep neural nets: the intuition;
– Why depth helps;
– Neural nets can learn to preprocess;
– Visualization of a deep embedding;
– The challenges to deep learning in the past + modern deep learning;
– Deep learning architectures;
– Convolution;
– Evolution of different components: ResNet, etc.
– Regularization in deep learning: early stopping, dropout, BatchNorm, ...
– Popular tricks:
– Augmentation;
– Transfer learning;
– Label smoothing;
– ...
