Deep Learning, key-points. Introduction to Neural Networks Quiz

Explore the essential principles of neural networks and deep learning, including their structure, training process, key functions, and regularization methods. This quiz provides foundational knowledge for machine learning and AI enthusiasts.

1. Neural Network Inspiration

   What is the main inspiration behind the design of artificial neural networks?

   1. Classical mechanics
   2. Biological neural systems
   3. Digital signal processing
   4. Electric circuits

   Explanation: Artificial neural networks are inspired by how biological neural systems, like the human brain, process and transmit information. Digital signal processing and electric circuits are related technologies but not the foundational inspiration. Classical mechanics pertains to physics and is unrelated to neural network design.

2. MLP Architecture Elements

   Which of the following is NOT typically considered a component of a multi-layer perceptron (MLP) architecture?

   1. Recurrent connections
   2. Input layer
   3. Hidden layer
   4. Weights

   Explanation: A multi-layer perceptron consists of an input layer, one or more hidden layers, and weights connecting the neurons. Recurrent connections, which allow feedback loops, are characteristic of recurrent neural networks, not standard MLPs.

3. Activation Functions in Deep Learning

   Why are activation functions such as ReLU often preferred over sigmoid or tanh in deep neural networks?

   1. They help mitigate the vanishing gradient problem
   2. They always produce binary outputs
   3. They increase training data size
   4. They eliminate the need for weights

   Explanation: ReLU activation functions are frequently used in deep neural networks because they help address the vanishing gradient problem, allowing gradients to propagate more effectively during training. They do not affect training data size or weights, and they produce non-binary outputs.

4. Regularization Techniques

   Which technique randomly deactivates a subset of neurons during training to enhance model robustness?

   1. Gradient boosting
   2. Dropout
   3. Batch normalization
   4. Cross-validation

   Explanation: Dropout randomly deactivates a subset of neurons during each training iteration to reduce overfitting and promote model robustness. Gradient boosting and cross-validation are different machine learning techniques, while batch normalization is used for stabilizing learning, not randomly dropping neurons.

5. Neural Network Training Process

   What is the primary role of backpropagation in training neural networks?

   1. To generate new training data
   2. To choose the activation function
   3. To normalize input data
   4. To compute gradients for updating weights

   Explanation: Backpropagation is used to compute gradients of the loss function with respect to each weight for optimizing the network during training. It does not generate data, select activation functions, or normalize input data.