ML Chapter 04

What is the goal of training a neural network?

• To collect large amounts of data
• To manually set weights
• To find the best weight parameters using data
• To reduce the number of neurons

What is a loss function used for?

• To visualize predictions
• To count training epochs
• To measure prediction error
• To update the learning rate

Which of the following is a type of loss function?

• Mean squared error
• Batch norm
• Dropout
• Adam

What is one-hot encoding?

• Encoding where all values are zero
• Encoding with multiple ones
• Encoding where only one value is 1, others are 0
• A form of encryption

Why is accuracy not used directly to train neural networks?

• It requires GPU support
• It doesn't change smoothly
• It increases loss
• It's too complex to compute

What is mini-batch learning?

• Training with the entire dataset
• Using random small portions of data for training
• Testing only one data point
• Pre-training with fake data

Why use mini-batches instead of full datasets?

• They increase model size
• They provide overfitting
• They reduce memory and computation load
• They lower accuracy

What is the purpose of numerical differentiation?

• To draw plots
• To manually set weights
• To approximate gradients
• To improve activation

What problem can occur with numerical differentiation?

• Vanishing gradient
• Overflow
• Rounding error
• Incorrect dataset

What is a gradient in neural networks?

• A random update
• A measurement of training speed
• A vector of partial derivatives
• An accuracy graph