Machine Learning and Deep Learning Quiz

What is the main distinction between Machine Learning (ML) and traditional programming?

• ML systems require no data to function

• ML systems learn patterns from examples rather than following only fixed rules

• ML systems never need human supervision

• Traditional programming always uses neural networks

Which of the following is NOT a common application area for Deep Learning?

• Image and speech recognition

• Simple rule-based arithmetic

• Natural language processing

• Autonomous driving

In Supervised Learning, the primary requirement for training is:

• Having a large dataset without any labels

• Providing explicit rules to the model

• Having labeled examples to learn from

• Using a reward and penalty mechanism

Which statement best describes a neuron in an artificial neural network?

• A function that randomly generates outputs regardless of the input

• A function that applies weights, bias, and an activation function to generate an output

• A part of the network that stores training examples

• A purely biological component mimicking the human brain

Which neural network architecture is most commonly used for image recognition tasks?

• Recurrent Neural Network (RNN)

• Multi-Layer Perceptron (MLP)

• Convolutional Neural Network (CNN)

• Transformer

Why are activation functions crucial in neural networks?

• They convert a regression problem into a classification problem automatically

• They introduce nonlinearity, enabling networks to learn complex patterns

• They ensure the network only learns linear relationships

• They guarantee the network never overfits

Which activation function is best avoided in hidden layers due to strong vanishing gradients?

• ReLU

• Sigmoid

• Tanh

• Softmax

What is the primary role of the bias term in a neuron’s computation?

• To reduce the number of parameters

• To introduce a nonlinearity in the activation function

• To shift the activation function output left or right

• To serve as a dropout rate

In Reinforcement Learning, the model learns by:

• Observing many labeled images

• Maximizing a reward through trial-and-error interactions

• Clustering unlabeled data into groups

• Backpropagation without any data

Which of the following is a key advantage of Transformers over RNNs for sequence data?

• They do not need attention mechanisms

• They can process sequences in parallel, speeding up training

• They rely exclusively on recurrence to handle long sequences

• They are only suitable for small datasets