1. What is the fundamental unit of computation in a neural network inspired by the human brain?
a) Perceptron
b) Neuron
c) Node
d) Synapse
Answer: b) Neuron
Explanation: The McCulloch-Pitts neuron model, proposed in the 1940s, serves as the fundamental unit of computation in neural networks, inspired by the biological neurons in the human brain.
2. Which logic was employed by the McCulloch-Pitts neuron for decision making?
a) Linear Logic
b) Fuzzy Logic
c) Thresholding Logic
d) Boolean Logic
Answer: c) Thresholding Logic
Explanation: The McCulloch-Pitts neuron employs thresholding logic, where the output is activated only if the input exceeds a certain threshold.
3. Which of the following is NOT an activation function used in neural networks?
a) ReLU
b) Sigmoid
c) Tanh
d) Heaviside Step Function
Answer: d) Heaviside Step Function
Explanation: The Heaviside step function is a discontinuous function often used in mathematics but not commonly employed as an activation function in neural networks.
4. Which optimization algorithm aims to minimize the loss function by iteratively adjusting the model parameters in the direction of the steepest descent of the loss surface?
a) Genetic Algorithm
b) Gradient Descent
c) Ant Colony Optimization
d) Simulated Annealing
Answer: b) Gradient Descent
Explanation: Gradient Descent iteratively adjusts model parameters in the direction of the steepest descent of the loss surface to minimize the loss function.
5. Which variant of Gradient Descent incorporates a momentum term to accelerate convergence and dampen oscillations?
a) Momentum Based Gradient Descent
b) Stochastic Gradient Descent
c) AdaGrad
d) RMSProp
Answer: a) Momentum Based Gradient Descent
Explanation: Momentum Based Gradient Descent incorporates a momentum term to accelerate convergence and dampen oscillations, especially in regions with high curvature.
6. Which variant of Gradient Descent modifies the momentum update to consider the “lookahead” position when computing the gradient?
a) Momentum Based Gradient Descent
b) Stochastic Gradient Descent
c) Nesterov Accelerated Gradient Descent
d) AdaGrad
Answer: c) Nesterov Accelerated Gradient Descent
Explanation: Nesterov Accelerated Gradient Descent modifies the momentum update to consider the “lookahead” position when computing the gradient, improving convergence.
7. Which variant of Gradient Descent updates the model parameters using a single training example per iteration?
a) Batch Gradient Descent
b) Mini-batch Gradient Descent
c) Stochastic Gradient Descent
d) AdaGrad
Answer: c) Stochastic Gradient Descent
Explanation: Stochastic Gradient Descent updates the model parameters using a single training example per iteration, making it faster but more noisy compared to batch methods.
8. Which optimization algorithm adapts the learning rate for each parameter based on the historical gradients for that parameter?
a) Momentum Based Gradient Descent
b) AdaGrad
c) RMSProp
d) Adam
Answer: b) AdaGrad
Explanation: AdaGrad adapts the learning rate for each parameter based on the historical gradients for that parameter, effectively scaling the learning rate.
9. Which optimization algorithm addresses the diminishing learning rate problem by accumulating squared gradients for each parameter and adapting the learning rate accordingly?
a) Momentum Based Gradient Descent
b) AdaGrad
c) RMSProp
d) Adam
Answer: c) RMSProp
Explanation: RMSProp addresses the diminishing learning rate problem by accumulating squared gradients for each parameter and adapting the learning rate accordingly.
10. Which optimization algorithm combines the advantages of both AdaGrad and RMSProp by incorporating an exponentially decaying average of past gradients and squared gradients?
a) Momentum Based Gradient Descent
b) AdaGrad
c) RMSProp
d) Adam
Answer: d) Adam
Explanation: Adam combines the advantages of both AdaGrad and RMSProp by incorporating an exponentially decaying average of past gradients and squared gradients.
11. Which mathematical technique is commonly used to analyze the properties of neural network weight matrices?
a) Singular Value Decomposition
b) Principal Component Analysis
c) Eigenvalue Decomposition
d) Fourier Transform
Answer: c) Eigenvalue Decomposition
Explanation: Eigenvalue Decomposition is commonly used to analyze the properties of neural network weight matrices, especially in understanding stability and convergence properties.
12. Which type of neural network is specifically designed to handle sequential data by maintaining state information?
a) Convolutional Neural Network
b) Recurrent Neural Network
c) Feedforward Neural Network
d) Radial Basis Function Network
Answer: b) Recurrent Neural Network
Explanation: Recurrent Neural Networks (RNNs) are designed to handle sequential data by maintaining state information and have connections that form directed cycles.
13. What is the technique used to train recurrent neural networks over a sequence of time steps by unfolding them into feedforward networks?
a) Backpropagation through Layers (BPL)
b) Backpropagation through Time (BPTT)
c) Gradient Descent through Time (GDTT)
d) Temporal Backpropagation (TBP)
Answer: b) Backpropagation through Time (BPTT)
Explanation: Backpropagation through Time (BPTT) is the technique used to train recurrent neural networks over a sequence of time steps by unfolding them into feedforward networks.
14. What is the problem encountered during backpropagation through time where gradients either vanish or explode as they are propagated backward in time?
a) Gradient Descent Issue
b) Vanishing and Exploding Gradients
c) Backpropagation Overfitting
d) Convergence Divergence
Answer: b) Vanishing and Exploding Gradients
Explanation: Vanishing and Exploding Gradients are the problems encountered during backpropagation through time where gradients either diminish to zero or grow exponentially as they are propagated backward in time.
15. Which technique limits the extent of backpropagation through time to mitigate the vanishing and exploding gradients problem?
a) Limited Memory BPTT
b) Truncated BPTT
c) Shortened BPTT
d) Minimized BPTT
Answer: b) Truncated BPTT
Explanation: Truncated BPTT limits the extent of backpropagation through time, thereby mitigating the vanishing and exploding gradients problem.
16. Which type of recurrent neural network architecture is designed to alleviate the vanishing gradient problem by incorporating gating mechanisms?
a) LSTM (Long Short-Term Memory)
b) GRU (Gated Recurrent Unit)
c) Elman Network
d) Jordan Network
Answer: a) LSTM (Long Short-Term Memory)
Explanation: LSTM (Long Short-Term Memory) is designed to alleviate the vanishing gradient problem in recurrent neural networks by incorporating gating mechanisms to regulate the flow of information.
17. Which recurrent neural network architecture simplifies the LSTM architecture by merging the forget and input gates into a single “update gate”?
a) LSTM (Long Short-Term Memory)
b) GRU (Gated Recurrent Unit)
c) Elman Network
d) Jordan Network
Answer: b) GRU (Gated Recurrent Unit)
Explanation: GRU (Gated Recurrent Unit) simplifies the LSTM architecture by merging the forget and input gates into a single “update gate,” reducing the number of parameters.
18. Which type of neural network architecture is commonly used for tasks such as machine translation and image captioning, consisting of an encoder and a decoder?
a) Convolutional Neural Network
b) Recurrent Neural Network
c) Transformer
d) Autoencoder
Answer: c) Transformer
Explanation: Transformer architecture is commonly used for tasks such as machine translation and image captioning, consisting of an encoder and a decoder with attention mechanisms.
19. What is the mechanism in neural networks that allows the model to selectively focus on different parts of the input when making predictions?
a) Weight Pruning
b) Dropout
c) Batch Normalization
d) Attention Mechanism
Answer: d) Attention Mechanism
Explanation: Attention Mechanism allows neural networks to selectively focus on different parts of the input when making predictions, improving performance in tasks like machine translation and image captioning.
20. Which attention mechanism variant is specifically designed to handle input sequences of variable lengths?
a) Global Attention
b) Local Attention
c) Self-Attention
d) Hierarchical Attention
Answer: c) Self-Attention
Explanation: Self-Attention, also known as intra-attention, is specifically designed to handle input sequences of variable lengths by attending to different positions of the sequence.
21. What is the process of extending attention mechanisms to work with images called?
a) Image Attention
b) Visual Attention
c) Spatial Attention
d) Image Localization
Answer: c) Spatial Attention
Explanation: Spatial Attention is the process of extending attention mechanisms to work with images by selectively focusing on different spatial regions.
22. Which of the following is NOT an application of attention mechanisms in neural networks?
a) Machine Translation
b) Image Classification
c) Image Captioning
d) Speech Recognition
Answer: b) Image Classification
Explanation: While attention mechanisms are widely used in tasks like machine translation, image captioning, and speech recognition, they are not commonly applied directly to image classification tasks.
23. Which of the following is NOT a component of a recurrent neural network?
a) Input Layer
b) Hidden Layer
c) Output Layer
d) Pooling Layer
Answer: d) Pooling Layer
Explanation: Pooling layers are commonly used in convolutional neural networks but are not typically included in recurrent neural network architectures.
24. Which type of neural network architecture is well-suited for handling sequential data and has memory elements that store information about previous states?
a) Convolutional Neural Network
b) Recurrent Neural Network
c) Feedforward Neural Network
d) Radial Basis Function Network
Answer: b) Recurrent Neural Network
Explanation: Recurrent Neural Networks (RNNs) are well-suited for handling sequential data and have memory elements that store information about previous states.
25. Which of the following is NOT a variant of the LSTM (Long Short-Term Memory) architecture?
a) Peephole LSTM
b) Bidirectional LSTM
c) Deep LSTM
d) Residual LSTM
Answer: d) Residual LSTM
Explanation: Residual LSTM is not a standard variant of the LSTM architecture. Peephole LSTM, Bidirectional LSTM, and Deep LSTM are commonly recognized variants.
26. What is the primary function of the encoder in an encoder-decoder model?
a) Generating Outputs
b) Processing Input Data
c) Calculating Loss
d) Storing Memory
Answer: b) Processing Input Data
Explanation: The encoder in an encoder-decoder model processes input data and typically produces a compact representation, which is then used by the decoder to generate outputs.
27. Which technique is used to alleviate the computational burden of attending to the entire input sequence in sequence-to-sequence models?
a) Global Attention
b) Local Attention
c) Self-Attention
d) Hierarchical Attention
Answer: b) Local Attention
Explanation: Local Attention is used to alleviate the computational burden of attending to the entire input sequence by focusing only on a subset of the input at each decoding step.
28. Which type of neural network architecture is designed to compress input sequences into a fixed-size vector representation?
a) Autoencoder
b) Convolutional Neural Network
c) Recurrent Neural Network
d) Transformer
Answer: a) Autoencoder
Explanation: Autoencoders are designed to compress input sequences into a fixed-size vector representation through an encoder network and then reconstruct the input from this representation using a decoder network.
29. Which variant of the encoder-decoder model incorporates attention mechanisms to improve performance in tasks such as machine translation?
a) LSTM-Attention
b) Transformer
c) GRU-Attention
d) Bi-LSTM
Answer: b) Transformer
Explanation: Transformer is a variant of the encoder-decoder model that incorporates attention mechanisms and has shown remarkable performance improvements in tasks such as machine translation.
30. Which technique is used to handle long-range dependencies in sequences by capturing relationships between distant elements more effectively?
a) Self-Attention
b) LSTM
c) Skip Connections
d) Positional Encoding
Answer: a) Self-Attention
Explanation: Self-Attention is used to handle long-range dependencies in sequences by capturing relationships between distant elements more effectively, allowing the model to attend to relevant information regardless of distance.