WorksheetsArtificial Neural Networks Worksheet Questions (Grade 13)
Total questions: 50
Worksheet time: 25mins
Which of the following is a characteristic of Artificial Neural Networks (ANN)?
Linear processing only
Distributed and parallel processing
Rule-based computation
Symbolic reasoning
What is the primary function of dendrites in a biological neuron?
Send signals
Receive input signals
Store information
Generate impulses
An artificial neuron is defined as:
A biological cell
A mathematical model of a neuron
A rule-based system
A digital switch
Which activation function is typically used in multiclass classification?
Sigmoid
ReLU
Tanh
Softmax
A continuous perceptron network uses:
Binary outputs
Continuous activation functions
Step function only
No weights
In a Hopfield network, the weights are:
Random
Asymmetric
Symmetric
Always zero
One learning difficulty in multilayer feedforward networks is:
Overfitting
Low storage
High learning rate stability
Zero training time
The main objective of the backpropagation algorithm is to:
Increase loss
Minimize error by updating weights
Increase the number of neurons
Initialize weights randomly
Data normalization techniques include:
Min–Max scaling and Z-score normalization
Fourier transform
Gradient scaling
Batch clipping
Deep Belief Networks (DBN) use which learning method?
Reinforcement learning
Unsupervised pretraining
Genetic learning
Hebbian learning
A biological neuron consists of:
Dendrites, axon, soma
Wires, sensors, chips
Weights, bias, activation
Inputs, outputs, loss
A simple ANN architecture includes:
Input, hidden, output layers
Processor and memory
Encoder and decoder
Sensor and actuator
A limitation of reinforcement learning is:
Requires labeled data
No interaction with environment
No rewards
Easy convergence
A single-layer feedforward neural network has:
One input and one output layer
Multiple hidden layers
Feedback connections
Recurrent units
If the learning rate of a perceptron is too large:
Learning becomes unstable
Convergence is faster
The weights freeze
Accuracy improves
The generalized delta rule is used in:
Perceptron learning
Backpropagation learning
Reinforcement learning
Genetic algorithms
Backpropagation uses which type of learning strategy?
Supervised learning
Unsupervised learning
Semi-supervised learning
Reinforcement learning
Network normalization methods include:
Batch normalization, Layer normalization
Dropout, Early stopping
Regularization only
Max pooling
Artificial vs Biological Neural Networks differ because:
ANNs are digital; biological are analog
Both process chemically
Both are hardware-based
ANNs use DNA-based memory
Total parameters in a fully connected network with 3 input, 5 hidden, and 10 output neurons:
15
80
40
60
Steps in backpropagation include:
Forward pass → Error calculation → Backward weight update
Weight freeze → Output reset
Only error calculation
Random updates
Deep feedforward neural networks contain:
Multiple hidden layers
Only one output neuron
Recurrent loops
No bias terms
Biological vs Artificial Neuron difference:
Biological uses electrochemical signals; Artificial uses mathematical functions
Both are identical
Artificial uses chemical signals
Biological uses digital pulses
Hopfield network energy function minimizes:
System error
Output variance
Entropy
Gradient
Purpose of backpropagation:
Update weights to reduce error
Freeze weights
Increase bias
Skip training
Machine Learning vs Deep Learning difference:
Deep learning uses neural networks with multiple layers
Machine learning uses deep layers
Deep learning uses rules
ML does not learn
Structure of Deep Belief Network includes:
Stacked Restricted Boltzmann Machines
Feedforward perceptrons
CNN layers
RNN cells
The sigmoid function output range is:
(-1, 1)
(0, 1)
(-∞, ∞)
(0, ∞)
The tanh activation function output range is:
(0, 1)
(-1, 1)
(-∞, ∞)
(0, ∞)
ReLU stands for:
Rectified Linear Unit
Relative Linear Unit
Reduced Logic Unit
Recursive Layer Unit
Gradient vanishing occurs in:
Deep networks with sigmoid/tanh
Shallow networks only
CNNs always
RNNs with ReLU
In supervised learning, training data contains:
Only inputs
Inputs with labeled outputs
Only outputs
Random signals
Dropout is used to:
Prevent overfitting
Reduce training data
Increase bias
Increase learning rate
Batch normalization helps to:
Stabilize and speed up training
Reduce batch size
Eliminate gradients
Decrease hidden units
Learning rate controls:
Step size in weight updates
Number of neurons
Network depth
Output value
Hebbian learning principle states:
Neurons that fire together wire together
Neurons that rest together stay together
Random firing
Decrease in weights
The main objective of normalization:
Scale inputs for faster convergence
Change feature relationships
Randomize data
Reduce sample size
The bias in a neuron helps to:
Shift the activation function
Scale inputs
Increase learning rate
Reduce weights
Which network uses energy minimization principle?
Hopfield Network
CNN
RNN
Perceptron
The softmax function outputs:
Probabilities that sum to one
Binary outputs
Negative values
Random scores
Epoch in neural network training refers to:
One complete pass through training data
A single batch update
A test cycle
Random initialization
Overfitting occurs when:
Model learns noise in training data
Model underperforms on training data
Model generalizes well
Learning rate is small
Which of the following is a deep learning framework?
TensorFlow
MATLAB
Excel
Power BI
Reinforcement learning learns by:
Reward and punishment signals
Predefined labels
Random guessing
Manual updates
In a discrete Hopfield network, the weight matrix is typically...
Asymmetric
Symmetric (Wij=Wji)
Trained using backpropagation
A diagonal matrix
Synaptic transmission, the communication between neurons, is primarily mediated by chemicals called...
Neurotransmitters
Enzymes
Hormones
Electrolytes
What are the two main 'passes' of the backpropagation algorithm?
Input Pass and Output Pass
Normalization Pass and Training Pass
Weight Initialization and Error Calculation
Forward Pass and Backward Pass
The Hopfield network is a form of...?
Feedforward Neural Network
Recurrent Neural Network
Convolutional Neural Network
Autoencoder
What is the primary purpose of the 'activation function' in an artificial neuron?
To sum the weighted inputs
To normalize the input data
To introduce non-linearity into the network
To store the weight value
A fully connected deep feedforward network is characterized by information flowing...
Only in one direction, from input to output, through the layers
In cycles, allowing for memory
Only between neurons in the same layer
From the output layer back to the input layer
