ASSIGNMENT

1) What is deep learning, and how does it contrast with other machine learning
algorithms?
2) How do you ensure youre not overfitting with a model?
3) Define an auto-encoder and its purpose. What role do auto-encoder play in machine
learning
4) Whats the kernel trick and how is it useful?
5) Answer the following questions in context of convolution neural networks
a) Explain why small convolution filters are used in CNN.
b) What is the use of 1 X 1 convolutions.
c) Explain how the size of a feature map between two convolutional layers can be
reduced just like pooling but without using pooling
6) Explain how an encoder-decoder architecture is used for machine translation (both during
training and during inference). Furthermore, elaborate on how attention mechanism works.
7) Explain how unrolling through time is related to weight sharing in convolution
neural network.
8) For the three sentences: It was the best of times, It was the
worst of times and It was the age of wisdom map the content
onto a skip-thought model. Sketch the model with circles to represent RNN nodes such
as LSTMs.
9) Specify how convolution of a given image is computed. Assume the input is an image
I of size H W with C channels, the kernel K has size N M, the stride is T S, the
operation performed is in fact cross-correlation (as usual in convolutional neural
networks) and that O output channels are computed. Spell out the computations for both
SAME and VALID padding schemes.
Compare Boltzmann Machine, Restricted Boltzmann Machine and Deep Belief Network.