Datasets

Question 1

Hypertuning

Answer 1

This is the process of tuning and adjusting hyperparameters of the neural network to optimize for its performance

Question 2

3 types of datasets

Answer 2

Training

Validation

Testing

Question 3

Training Dataset

Answer 3

Used to train a neural network to produce a desired output. Contains input data with known output (correct answer) which the NN can use for comparison later to calculate loss.

Question 4

Validation Dataset

Answer 4

Used for hypertuning.

Detects adjustments needed that will improve the performance of the NN.

Question 5

Testing dataset

Answer 5

Used to evaluate the performance of the NN. It also contains the “Correct prediction/answer” however testing data must not overlap with the training data. 2 different things.

Question 6

Hyperparameters what are they and give 2 of them.

Answer 6

These are aspects of the architecture of the NN that can be adjusted to optimize performance of the network.

1. Number of layers
   - More layers can lead to more precision
   - However, more layers can lead to vanishing gradient problem
   - It also may require more memory and processing power
2. Learning rate
   - A faster learning rate can lead to the NN learning to produce the correct output more quickly
   - A faster learning rate may however lead to the NN to stop learning too early, ultimately creating an inaccurate network

Question 7

Learning Rate

Answer 7

How dramatically weights are changed in response to the calculated gradients