What is a Neural Network primarily inspired by?

• The internet

• The human brain

• A flowchart

• The periodic table

What is the primary function of a Perceptron in a neural network?

• It acts as the output layer of a neural network

• It mimics the process of a biological neuron

• It prevents overfitting in deep learning models

• It is used for data visualization

Which problem occurs in deep neural networks due to backpropagation causing earlier layers to receive very small error signals?

• Overfitting

• Data leakage

• Vanishing gradient problem

• Over-saturation

What differentiates deep learning from traditional machine learning?

• Deep learning requires no training data

• Deep learning relies on predefined feature extraction

• Deep learning automatically extracts features from raw data

• Traditional machine learning uses more layers than deep learning

Which of the following is NOT an application of deep learning?

• Image classification

• Speech recognition

• Email filtering

• Machine translation

Which activation function is commonly used to help mitigate the vanishing gradient problem?

• Sigmoid

• Tanh

• Softmax

• Rectified Linear Unit (ReLU)

What is the advantage of using PyTorch over other deep learning frameworks?

• It does not require coding knowledge

• It uses dynamic computation graphs for flexibility

• It is the only framework that supports deep learning

• It is built only for mobile applications

What is the role of Batch Normalization in deep learning?

• It slows down the training process

• It prevents neural networks from memorizing training data

• It stabilizes the learning process and reduces training time

• It removes all errors from a deep learning model

Why are GPUs preferred over CPUs for deep learning?

• GPUs are cheaper than CPUs

• GPUs can perform complex matrix operations much faster

• CPUs are not capable of deep learning

• GPUs require less memory

What is the main benefit of using deep learning frameworks like TensorFlow and PyTorch?

• They eliminate the need for labeled data

• They simplify the implementation of deep learning models

• They do not require GPU support

• They replace the need for neural networks