Neural Networks Explained quiz

Question 1

What is a neural network?

Accepted Answer

An AI system made of layers of connected nodes that learn from examples

Answer

A network of computers connected to the internet

Answer

A type of database organized in a tree structure

Answer

A computer chip designed to look like a brain

Question 2

What are 'layers' in a neural network?

Accepted Answer

Layers organize neurons into input, hidden, and output sections, each learning different features

Answer

A layer is a single AI model stacked on top of another

Answer

Layers are only used in image-recognition networks

Answer

A layer is a database table connected to the network

Question 3

What is 'bias' in a neural network (not data bias)?

Accepted Answer

An extra number added to a neuron's input sum to shift its activation threshold

Answer

Unfair preferences the network develops from training data

Answer

The number of errors in the training dataset

Answer

A penalty applied to overfitting networks

Question 4

What is the 'ReLU' activation function?

Accepted Answer

A function that outputs the input if positive, or zero if negative

Answer

A function that outputs random numbers between 0 and 1

Answer

A function that always outputs exactly 0 or 1

Answer

A function that reverses negative inputs

Question 5

What is 'backpropagation'?

Accepted Answer

Working backward through the network using calculus to adjust weights to reduce prediction error

Answer

Propagating data from output layer back to input for verification

Answer

Removing neurons that are not contributing to predictions

Answer

Reversing the order of training data for efficiency

Question 6

What is a 'CNN' (Convolutional Neural Network)?

Accepted Answer

A neural network using sliding filters to detect patterns in images

Answer

A neural network that only works with text data

Answer

A neural network for predicting continuous values

Answer

A network for connecting multiple AI models together

Question 7

What is 'LSTM' and how does it improve on basic RNNs?

Accepted Answer

An improved RNN with gates that control what to remember or forget over long sequences

Answer

A very large RNN with thousands of neurons

Answer

A type of CNN used for 3D images

Answer

A special type of transformer layer

Question 8

What is 'batch normalization'?

Accepted Answer

Normalizing layer inputs within each training batch for stability and faster training

Answer

Splitting training data into random batches

Answer

A technique for normalizing the final output of a model

Answer

Removing abnormal examples from training batches

Question 9

What is a 'residual network' (ResNet)?

Accepted Answer

A network with skip connections that allows very deep architectures by bypassing vanishing gradients

Answer

A network that reuses the same weights for every layer

Answer

A network that resets its weights after each training epoch

Answer

A recurrent network for residual data processing

Question 10

What does 'epoch' mean in neural network training?

Accepted Answer

One complete pass through the entire training dataset during model training

Answer

The time it takes to train one neuron

Answer

A type of evaluation metric for neural networks

Answer

The number of layers in a deep neural network

Question 11

What is an 'embedding layer' in a neural network?

Accepted Answer

A layer that maps categorical inputs like words to dense numerical vectors capturing meaning

Answer

A layer that reduces the depth of the network

Answer

A layer that processes images into numbers

Answer

A layer that outputs final classification probabilities

Question 12

What is 'overfitting' in a neural network?

Accepted Answer

When a network memorizes training data too precisely and fails on new examples

Answer

When the network trains for too many hours without stopping

Answer

When the network has too few neurons to learn anything

Answer

When training data is accidentally mixed with test data

Question 13

What is 'attention' in the context of neural networks?

Accepted Answer

A mechanism letting the network focus on the most relevant parts of input for each output

Answer

A mechanism that focuses only on the last word in a sequence

Answer

A technique for reducing the number of neurons in a layer

Answer

A method for speeding up model inference

Question 14

What is 'pooling' in convolutional neural networks?

Accepted Answer

Reducing feature map size by taking the max or average of small regions

Answer

Combining multiple neural networks into one pool

Answer

Adding more neurons to a specific layer

Answer

A technique for initializing neural network weights

Question 15

What is a 'hyperparameter' in neural network design?

Accepted Answer

Settings chosen before training — like learning rate and number of layers — not learned by the network

Answer

Parameters learned automatically during training

Answer

Parameters stored in the final neural network layer

Answer

A special type of weight used only in transformer models

Question 16

What is 'parameter count' and why does it matter?

Accepted Answer

The total number of weights and biases in a neural network

Answer

The number of features in the training dataset

Answer

The number of training examples used

Answer

The number of predictions the model can make per second

Question 17

What is the 'exploding gradient' problem?

Accepted Answer

When gradients grow exponentially backward, causing massive unstable weight updates

Answer

When gradient descent moves too slowly toward the minimum

Answer

When the learning rate is too low to train efficiently

Answer

When a neural network generates explosive outputs for certain inputs

Question 18

What is 'weight initialization' and why does it matter?

Accepted Answer

Setting starting weight values carefully so training begins in a healthy range

Answer

Resetting weights to their best values after training

Answer

Downloading pre-trained weights from the internet

Answer

The process of loading training data into the network

Question 19

What is a 'fully connected layer' (dense layer)?

Accepted Answer

A layer where every neuron connects to every neuron in the previous layer

Answer

A layer where each neuron connects to only one neuron in the next layer

Answer

A layer used only for image processing

Answer

A layer with no trainable parameters

Question 20

What is 'AlexNet' and why was it historically important?

Accepted Answer

A deep CNN that won ImageNet 2012 by a huge margin, launching the modern deep learning era

Answer

The first language model ever created

Answer

The first neural network to speak human language

Answer

A hardware chip designed for neural network inference

Question 21

What role do GPUs play in neural network training?

Accepted Answer

GPUs do massive parallel matrix math 10–100x faster than CPUs, making deep learning practical

Answer

GPUs are needed only to display the neural network diagram on screen

Answer

GPUs are used only to store neural network weights

Answer

CPUs are always faster than GPUs for neural network training

Question 22

What is 'batch size' and how does it affect training?

Accepted Answer

The number of examples processed before updating model weights — affecting training stability and speed

Answer

The size of the training dataset in megabytes

Answer

The number of neural network layers in the model

Answer

The amount of GPU memory allocated for one training run

Question 23

What is 'weight tying' in neural network language models?

Accepted Answer

Sharing weights between the input embedding and output projection matrices to reduce model size

Answer

Securing neural network weights from unauthorized access

Answer

A technique for preventing weight updates during fine-tuning

Answer

Connecting neural network layers with physical cables in hardware

Question 24

What is 'data parallelism' in deep learning training?

Accepted Answer

Splitting training batches across multiple GPUs, each running a full model copy, then synchronizing gradients

Answer

Running two different models on the same data simultaneously

Answer

A method for loading training data faster from disk

Answer

Running the same model on both CPU and GPU at once

Question 25

What is 'neural scaling' and why does it matter?

Accepted Answer

Predictable performance improvements following power laws as model size, data, and compute increase

Answer

The process of resizing images to fit a neural network's input size

Answer

Making a neural network work on smaller devices by scaling it down

Answer

A technique for measuring neural network speed

Question 26

What is 'neural network interpretability'?

Accepted Answer

Research into what neural networks actually do internally — which neurons represent what concepts

Answer

Making neural networks respond faster to user queries

Answer

A legal requirement to explain all neural network decisions

Answer

Training neural networks on interpretable datasets only

Test yourself