Reference no: EM132519716
Machine Learning
Clearly explain how each answer has been produced and give good motivations for all calculations. In order to obtain a good grade on all intermediary steps in the method that you use to solve the problem should be carefully presented.
You may not use any machine learning software, or anything related to it of any kind for solving or helping you solve the exam problems
Q: Which problems indeeplearning are solved by centering and scaling the neuron activations resulting from a mini batch?
Q:The following problems are related to neural nets:
(a) : One of the simplest useful optimization algorithms is gradient descent Consider the function y = (x - 2)2 Use a learning rate of 0.1 to perform the first three gradient descent steps starting at x0 = 5.
(b) : In this dataset, the task is to predict if a person becomes a world class athlete based on three genetic mutations, namely ACTN3, CETP and LRP5. The following data has been collected.
|
Name
|
ACTN3
|
CETP
|
LRP5
|
World class athlete
|
|
John
|
yes
|
no
|
yes
|
yes
|
|
Mary
|
no
|
yes
|
no
|
no
|
|
Adam
|
yes
|
no
|
no
|
yes
|
|
Donald
|
yes
|
yes
|
no
|
no
|
|
Jill
|
no
|
no
|
no
|
no
|
Encode the mutation dataset so that it becomes suitable for neural nets. Assume that an output value less than or equal to zero means "no" and that an output greater than zero means "yes" .Give a neural net that only consists of one single ReLU node and that perfectly classifies the dataset. Carefully explain how you obtained the network and include all details such as the weights, of course.