Deep Learning MCQ: Deep learning is a branch of machine learning that enables computers to learn tasks independently, much like humans do. Deep neural networks have numerous hidden layers between the input and output layers, therefore the term “deep” in deep learning.
Deep learning models take instructions from how the human brain works and is structured. They are made up of interconnected nodes (neurons) that analyze and change data as they learn from big datasets through a procedure known as supervised learning.
Deep Learning MCQ With Revision Notes
Application of Deep Learning
Computer Vision:
- Image Classification: Deep learning is used Image classification algorithm to identify cats and dogs in pictures and other visual content.
- Object Detection: In surveillance systems and self-driving cars, it helps recognize and locate various things contained within an image or video stream.
- Facial Recognition: Deep learning may be used to identify and authenticate people using face characteristics, which are often utilized in security and access control.
- Image Generation: It is used for visual authenticity, entertainment, and design.
Natural Language Processing (NLP):
- Machine Translation: Deep learning models have greatly enhanced machine translation systems like Google Translate.
- Sentiment Analysis: It is helpful to identify the sentiment or emotional tone of information in order to assess customer comments and monitor social media.
- Chatbots and Virtual Assistants: Deep learning-based conversational AI, which is used by chatbots and virtual assistants like Siri and Alexa, allows these systems to comprehend and respond to human language.
- Text Generation: Models like GPT can provide language that is coherent and contextually suitable when used to generate chatbot responses and content.
Speech Recognition:
- Voice Assistants: Deep learning is essential for voice assistants like Apple’s Siri and Amazon Echo.
- Transcription Services: It is employed in closed captioning and transcription services to convert spoken material into text.
- Speaker Verification: Deep learning in security applications can verify someone’s identity based on their speech.
Deep learning various types or architectures
- Feedforward Neural Network (FNN): The most basic type of neural network is the feed-forward neural network (FNN), which consists of an input layer, one or more hidden layers, and an output layer. Regression and classification are prominent uses for it.
- CNN (Convolutional Neural Network): Convolutional neural networks, or CNNs, are used for image-related tasks like object detection and image classification.
- RNN (Recurrent Neural Network): RNNs (Recurrent Neural Networks) are effective for sequential data tasks like speech and natural language processing.
- LSTM (Long Short-Term Memory): LSTM (Long Short-Term Memory) is a type of RNN designed specifically for simulating distant dependencies in sequences.
- GRU (Gated Recurrent Unit): Another RNN variation noted for its effectiveness in sequence modeling is the GRU (Gated Recurrent Unit).
- Autoencoder: Used for data compression and unsupervised learning applications.
- GAN (Generative Adversarial Network): Used to provide accurate data, including text and photos.
- Transformer: This model, which is employed in models like BERT and GPT, revolutionized NLP.
- Siamese Network: Used frequently in tasks like facial recognition for learning about similarity and dissimilarity.
- Capsule Network (CapsNet): Created for analyzing hierarchical data.
- Deep Reinforcement Learning: This method of sequential decision-making combines deep learning and reinforcement learning.
Advantages of Deep Learning:
- High Performance: Produces excellent results across a variety of jobs.
- Feature Learning: Features are automatically learned from data using feature learning.
- Scalability: Able to handle difficult tasks and large datasets.
- Generalization: Applying knowledge to fresh, unforeseen data is generalization.
- End-to-End Learning: Makes modeling from unprocessed data easier.
- Transfer Learning: Reuses pre-trained models.
- Continuous Improvement: Constantly advancing technology.
Deep Learning MCQ
- What is the primary goal of deep learning?
a. Data compression
b. Feature extraction
c. Learning patterns from data
d. Linear regression
- What kind of neural network is most frequently applied to the image classification?
a. RNN (Recurrent Neural Network)
b. CNN (Convolutional Neural Network)
c. FNN (Feedforward Neural Network)
d. LSTM (Long Short-Term Memory)
- What is the main benefit of deep neural networks over shallow neural networks?
a. Faster training
b. Simplicity of architecture
c. Ability to learn complex features
d. Lower memory usage
- What is the primary advantage of using mini-batch gradient descent over batch gradient descent?
a. Faster convergence
b. Lower memory usage
c. Guaranteed global minimum
d. Simplicity of implementation
- What is the main function of pooling layer in a convolutional neural network (CNN)?
a. To add non-linearity
b. To increase model complexity
c. To reduce the spatial dimensions of feature maps
d. To add noise to the input data
- What kind of layer is commonly used in neural networks to add non-linearity?
a. Fully Connected Layer
b. Pooling Layer
c. Convolutional Layer
d. Activation Layer
- What does “epoch” mean in deep learning?
a. The number of layers in a neural network
b. The learning rate of the optimizer
c. One complete pass through the entire training dataset
d. The type of activation function used
- In deep learning, what is the role of a loss function?
a. To measure the model’s prediction accuracy
b. To initialize model parameters
c. To calculate the gradients for optimization
d. To normalize input data
- Which deep learning system is known for its dynamic computation graph and was created by Facebook AI Research?
a. TensorFlow
b. PyTorch
c. Keras
d. Theano
- What is the main benefit of batch gradient descent in deep learning over stochastic gradient descent?
a. It converges faster to the optimal solution.
b. It reduces memory usage during training.
c. It provides a more accurate estimate of the gradient.
d. It requires less computation.
- Which term best defines the process of improving a deep learning model that has already been trained for a particular task using a smaller dataset?
a. Model initialization
b. Data augmentation
c. Transfer learning
d. Gradient descent
- What are the key advantages of dropout in neural networks?
a. It increases model complexity.
b. It accelerates training.
c. It introduces stochasticity to improve generalization.
d. It reduces the number of neurons in the network.
- When artificial neural networks (ANNs) first proposed?
a. 1943
b. 1956
c. 1962
d. 1989
- Who is considered the father of artificial neural networks?
a. Ilya Sutskever
b. David Rumelhart
c. Yann LeCun
d. Frank Rosenblatt
- How many layers are there in a typical feed-forward neural network?
a. 1
b. 2
c. 3
d. It can vary depending on the problem
- What function do max-pooling layers serve in a convolutional neural network (CNN)?
a. To reduce the spatial dimensions of feature maps
b. To increase the number of channels in feature maps
c. To add noise to the input data
d. To introduce non-linearity
- What is the primary role of a loss function in deep learning?
a. To calculate the accuracy of a model
b. To determine the learning rate during training
c. To measure the difference between predicted and actual values
d. To determine the number of hidden layers in a network
- What deep learning framework has a reputation for being flexible and simple to use for creating neural networks?
a. TensorFlow
b. PyTorch
c. Keras
d. Caffe
- Which optimization technique is frequently employed to train deep neural networks and dynamically modifies learning rates?
a. Stochastic Gradient Descent (SGd.
b. Adam
c. Random Search
d. Mean Squared Error (MSE)
- What does the term “backpropagation” in deep learning refer to __________.
a. The process of updating model parameters during training
b. The use of gradient descent for optimization
c. The forward pass in a neural network
d. The selection of activation functions
- Which activation function, which has non-linearity and smooth gradient features, is frequently utilized in hidden layers of deep neural networks?
a. Sigmoid
b. ReLU (Rectified Linear Unit)
c. Tanh (Hyperbolic Tangent)
d. Linear
- What is “overfitting” in deep learning?
a. The inability of the model to fit the training data
b. The model learning relevant features from the data
c. The model’s tendency to fit noise in the training data
d. The model’s high training accuracy
- What does dropout regularization in deep neural networks do?
a. To add noise to the data
b. To increase the model’s complexity
c. To reduce overfitting
d. To accelerate training
- Which of the following is a typical activation function for introducing non-linearity in deep learning?
a. Mean Squared Error (MSE)
b. Rectified Linear Unit (ReLU)
c. Principal Component Analysis (PCa.
d. k-Nearest Neighbors (k-NN)
- Why does dropout regularization exist in neural networks?
a. To increase the number of hidden layers
b. To reduce the number of neurons in each layer
c. To prevent overfitting
d. To improve convergence speed
- What does “pooling” in convolutional neural networks (CNNs) mean?
a. Combining multiple layers into one
b. Reducing the spatial dimensions of the feature maps
c. Increasing the number of filters in a convolutional layer
d. Changing the activation function
- What kind of neural network is usually used for tasks involving the processing of sequential data and natural language?
a. Convolutional Neural Network (CNN)
b. Recurrent Neural Network (RNN)
c. Multilayer Perceptron (MLP)
d. Self-Organizing Map (SOM)
- What is the main role of the softmax activation function in a neural network’s output layer?
a. To introduce non-linearity
b. To normalize the output into probability distribution
c. To prevent vanishing gradients
d. To reduce the dimensionality of the data
- What does “epoch” mean when used in a deep learning model?
a. The number of layers in the neural network
b. The learning rate of the model
c. One complete pass through the entire training dataset
d. The number of neurons in the output layer
- What is the expected outcome of choosing a batch size that comfortably fits the memory when optimizing neural networks with plenty of RAM?
a. It may lead to slower training due to increased computational complexity.
b. It may result in faster training due to reduced data transfers and computations.
c. It has no significant effect on training speed.
d. It increases memory consumption during training.
- What will happen if you have a straightforward MLP model with three neurons, inputs of 1, 2, and 3, weights of 4, 5, and 6, and a constant linear activation function with a value of 3?
a. 22
b. 31
c. 35
d. 42
- What are the size of the weight matrices between the hidden and output layers, as well as between the input and hidden layers, in a simple MLP with 8 neurons in the input layer, 5 neurons in the hidden layer, and 1 neuron in the output layer?
a. 5×8 for hidden to input, and 1×5 for output to hidden.
b. 8×5 for hidden to input, and 5×1 for output to hidden.
c. 5×1 for hidden to input, and 1×5 for output to hidden.
d. 8×1 for hidden to input, and 1×8 for output to hidden.
- In each epoch of training a Deep Learning model, which of the following training parameters is constant?
a. Learning rate.
b. Batch size.
c. Number of hidden layers.
d. Loss function.
- Does the use of max pooling in convolutional neural networks (CNNs) necessarily result in a decrease in the number of parameters?
a. Yes.
b. No.
- Is sentiment analysis often regarded as a many-to-one prediction task when performed using deep learning?
a. Yes.
b. No.
- What are the dimensions of the convoluted matrix produced if an input picture has been turned into a 28×28 matrix and a 7×7 kernel/filter is used with a stride of 1?
a. 22×22.
b. 28×28.
c. 20×20.
d. 21×21.
- What is the typical impact of 1-1 convolutions on the number of channels in a CNN?
a. It increases spatial dimensions.
b. It reduces the number of channels.
c. It is only applicable to fully connected layers.
d. It adds more parameters to the network.
- When forecasting probabilities for n classes so that the total probability across n classes equals 1, what activation function is typically employed in the output layer?
a. ReLU.
b. Sigmoid.
c. Tanh.
d. Softmax.
- What is the maximum number of connections from the input layer to the hidden layer in a neural network with 10 input nodes and 5 hidden nodes?
a. 5.
b. 15.
c. 50.
d. 25.
- How accurate are deep learning algorithms compared to machine learning techniques for classifying images?
a. Slightly more accurate.
b. Significantly more accurate.
c. Equally accurate.
d. Occasionally more accurate.