mi1_deep_learning_sustainability

I've made a serious attempt at solving the past exercise sheet

• False

• True

The pace of the course so far is...

• ok

• too fast, I can't follow

• too slow, stop rambling!

An Auto-encoder...

• consists of an encoder+decoder

• is used for cryptocurrency

• is a new video codec

• teaches itself new tricks

A greedy algorithm is greedy because...

• it's data hungry.

• optimizes locally before globally.

• it contains memory leaks.

• finds lots of local optima.

The "Deep" in Deep Learning refers to....

• a model pondering about the meaning of life.

• learning a hierarchical structure.

• slow computation time.

• 3-dimensional data. Length, width and depth.

What does Backpropagation do?

• Compute gradients

• update weights

• remove noise from the input

• Gradient descent.

Backpropagation is used for gradient computations in Deep Neural Networks.

• True

• False

• Isn't Backpropagation a form of bird watching?

There's no such thing as Unsupervised Deep Learning.

• True

• False

Deepening an MLP (more layers) causes no problems ever (True/False).

• True

• False

We change architecture or training processes for DNNs, because, among other reason, ...

• we're all architects and this is fun.

• simply adding more layers to an MLP can cause the gradient to vanish.

• we just recently learned, how convolution works and we wanna try it out.

• we change the model pipeline and are always able to extract extra features.