CAT-1
What is the basic unit of a neural network?
A neuron (or perceptron).
What is the first pass of data through a network called?
Forward propagation.
Which type of network is best for identifying objects in photos?
CNN
What is it called when a model is too focused on its training data?
OVERFITTING
Which hardware chip is most commonly used to train AI?
GPU
What does the "Deep" in Deep Learning stand for?
Having many hidden layers.
What algorithm is used to calculate errors moving backward?
Backpropagation.
Which type of network is best for language and sequences?
RNN
What is it called when a model fails to learn the basic patterns?
UNDERFITTING
What is the most common non-linear activation function?
RELU
What are the three main types of layers in a network?
Input, Hidden, and Output layers.
What function measures the difference between a prediction and the truth?
Loss Function
What is the modern architecture used for models like GPT?
TRANSFORMER
What technique prevents overfitting by "switching off" neurons?
drop out
Which activation function is used for binary (Yes/No) classification?
sigmoid
What is the name for the data used to teach a model?
Training data.
What is one full pass through the entire dataset called?
epoch
What do we call a network that can create new images or text?
GEN AI
What is the goal of "Optimization"?
to minimize the loss function
What is the name of Google's open-source AI library?
TensorFlow.
What are the values that a network adjusts to learn?
Weights and Biases.
What setting controls the size of the steps taken during training?
learning rate
What type of learning uses "rewards" to train an agent?
RL
What happens to gradients when they become too small to update weights?
Answer: Vanishing Gradient problem.
What is a multi-dimensional array of numbers in Deep Learning called?
A Tensor.