Understanding Neural Network MCQs

Class 12 AI(843) students, this one’s for you 🎯

This Understanding Neural Networks MCQ collection is packed with important, exam-focused questions based on the latest CBSE syllabus and board paper pattern. Every question is carefully selected to cover the key concepts of the chapter, so you can revise thoroughly without wasting time. Practice these MCQs, strengthen your concepts, and walk into your CBSE exam with confidence.

Q1. What is a network of neurons called?
a) Machine Learning
b) Neural Network
c) Decision Tree
d) Data Network

Q2. The word “neural” in Neural Network is derived from:
a) Neural data
b) Neuron of the human nervous system
c) Network algorithm
d) Natural language

Q3. What is the basic unit of the brain that processes and transmits information?
a) Synapse
b) Neuron
c) Layer
d) Weight

Q4. What is the primary advantage of artificial neural networks over traditional programming approaches?
a) They require manually coded rules for every edge case.
b) They can extract data features automatically without input from the programmer.
c) They execute faster than any basic conditional logic.
d) They do not require any mathematical computations during training.

Q5. Which of the following is NOT listed in the handbook as a common application of Artificial Neural Networks (ANN)?
a) Spam filtering
b) Facebook image tagging
c) E-shopping product recommendations
d) Manual database indexing

Q6. What famous search engine algorithm is mentioned as one of the best-known examples of a neural network?
a) Yahoo Search
b) Bing Search
c) Google’s search algorithm
d) DuckDuckGo Search

Q7. What are the three core layers present in every neural network architecture?
a) First layer, Middle layer, Last layer
b) Input layer, Hidden layer(s), Output layer
c) Initial layer, Processing layer, Terminal layer
d) Feature layer, Weight layer, Prediction layer

Q8. Where are hidden layers located within a neural network architecture?
a) Before the input layer
b) Between input and output layers
c) After the output layer
d) Outside the neural network

Q9. The output layer represents:
a) Input features
b) Target fields
c) Weights only
d) Training data only

Q10. An ANN with two or more hidden layers is called:
a) Perceptron
b) Deep Neural Network
c) Linear Network
d) Simple Network

Q11. The term “deep” in deep learning refers to:
a) Amount of data
b) Number of hidden layers
c) Number of inputs
d) Size of output

Q12. The process of training deep neural networks is called:
a) Machine coding
b) Deep Learning
c) Data mining
d) Data cleaning

Q13. What are the fundamental building blocks of a neural network called?
a) Biases
b) Neurons (or nodes)
c) Filters
d) Hyperparameters

Q14. What component of a neural network represents the strength of connections between neurons?
a) Bias
b) Weight
c) Learning Rule
d) Epoch

Q15. During the training process of a neural network, what are adjusted to minimize error?
a) Number of input features
b) Optimal weights and biases
c) Target labels
d) Dataset size

Q16. Neurons are also known as:
a) Features
b) Nodes
c) Targets
d) Epochs

Q17. During training, neural networks learn optimal weights to:
a) Increase the error
b) Minimize the error
c) Remove inputs
d) Increase noise

Q18. What is the main role of Activation Functions in a neural network?
a) To store dataset values on disk.
b) To act as decision-makers and decide whether a neuron should be activated.
c) To reduce the number of input nodes.
d) To calculate historical training time.

Q19. Activation functions add what quality to a neural network, enabling it to learn complex patterns in data?
a) Constant slopes
b) Linearities
c) Non-linearities
d) Random noise

Q20. Which of the following is NOT an activation function mentioned in the unit text?
a) Sigmoid Function
b) Tanh Function
c) ReLU (Rectified Linear Unit)
d) Polynomial Kernel Function

Q21. Activation functions help neural networks understand:
a) Only simple numbers
b) Complex patterns in data
c) File formats
d) CSV structures

Q22. Bias is:
a) A constant added to the weighted sum
b) A type of neuron
c) A dataset
d) A loss function

Q23. What are bias terms in a neural network?
a) Variable multipliers applied to target outputs
b) Constants added to the weighted sum before applying the activation function
c) Errors measured during backpropagation
d) The number of training epochs

Q24. Connections between neurons in a neural network represent biological:
a) Synapses
b) Axons
c) Hormones
d) Blood vessels

Q25. The learning rule specifies:
a) How weights and biases are adjusted
b) How data is collected
c) How files are stored
d) How layers are displayed

Q26. Which algorithm computes gradients and updates weights to minimize error?
a) Backpropagation
b) Classification
c) Regression
d) Clustering

Q27. The forward pass is known as:
a) Back Propagation
b) Forward Propagation
c) Gradient Descent
d) Deep Learning

Q28. During forward propagation:
a) Input data flows through network layers
b) Weights are deleted
c) Only errors are transmitted
d) Data flows from output to input

Q29. The difference resulting from comparison of prediction and actual target is called:
a) Weight
b) Loss/error
c) Feature
d) Epoch

Q30. Backpropagation adjusts weights based on:
a) Dataset name
b) Error or loss
c) Number of input fields
d) File size

Q31. Gradients in backpropagation are used to:
a) Update weights
b) Create datasets
c) Add input features
d) Remove layers

Q32. Which optimization algorithm is mentioned as an example used with gradients?
a) Gradient descent
b) Linear regression
c) Classification
d) Clustering

Q33. The basic calculation performed by a neuron can be represented as:
a) w × x + b
b) x ÷ w − b
c) x + w − b
d) w − x + b

Q34. If a neuron fires, its output is:
a) Deleted
b) Passed to the next layer
c) Converted into a CSV file
d) Used as a learning rate

Q35. A network where information moves from input towards output is called:
a) Feedforward network
b) Recurrent network
c) GAN
d) Perceptron only

Q36. The perceptron was created by:
a) Frank Rosenblatt
b) Adam
c) TensorFlow
d) Google

Q37. Frank Rosenblatt created the perceptron in:
a) 1945
b) 1958
c) 1968
d) 1985

Q38. A perceptron is useful for:
a) Binary classification
b) Only regression
c) Only image generation
d) Data visualization

Q39. A Feed Forward Neural Network allows data to flow:
a) In loops
b) Only from input to output
c) From output to input
d) Randomly

Q40. Feed Forward Neural Networks (FFNN) are also known as:
a) Recurrent Networks
b) Multi-Layer Perceptrons (MLPs)
c) Generative Adversarial Networks
d) Convolutional Networks

Q41. CNNs use ______ to extract features from images.
a) Filters
b) Epochs
c) Biases only
d) Loss functions

Q42. CNNs are particularly effective for:
a) Visual data
b) Only numerical tables
c) Audio files only
d) CSV storage

Q43. Which neural network type is specifically highlighted as efficient for handling noisy data?
a) Standard Perceptron
b) Feed Forward Neural Network (FFNN)
c) Generative Adversarial Network (GAN)
d) Single-node TLU

Q44. What component do Convolutional Neural Networks (CNNs) utilize to extract features from images?
a) Feedback loops
b) Filters
c) Discriminators
d) Recurrent cells

Q45. Which network is dominant in computer vision?
a) RNN
b) CNN
c) GAN only
d) Perceptron only

Q46. RNNs are designed for:
a) Sequential data
b) Only images
c) Only static data
d) Database management

Q47. What structural feature allows RNNs to allow information persistence across time steps?
a) 3D Filters
b) Feedback loops / feedback connections
c) Single TLU layers
d) Unidirectional flow

Q48. Which of the following is an application of RNNs?
a) Machine translation
b) Image filtering only
c) Spreadsheet formatting
d) File compression

Q49. How many neural networks make up a Generative Adversarial Network (GAN)?
a) One single network
b) Two neural networks trained simultaneously
c) Three separate networks
d) Four parallel networks

Q50. A GAN consists of:
a) Two generators
b) Generator and discriminator
c) Two discriminators
d) Input and output only

Q51. What is the specific role of the Generator in a GAN?
a) Evaluates images for authenticity.
b) Creates new data instances.
c) Calculates linear regression error.
d) Filters high frequencies.

Q52. What is the role of the Discriminator in a GAN?
a) Generates fake noise.
b) Evaluates data instances for authenticity (judges real vs. fake).
c) Converts Celsius to Fahrenheit.
d) Sets the learning rate.

Q53. GANs are primarily used for which type of learning paradigm?
a) Supervised Learning
b) Unsupervised Learning
c) Semi-supervised rule matching
d) Reinforcement learning only

Q54. GANs can be used to generate:
a) Realistic images and videos
b) Only spreadsheets
c) Only text files
d) Only numerical values

Q55. Neural networks can improve efficiency by:
a) Automating tasks
b) Removing all workers
c) Eliminating data
d) Preventing innovation

Q56. Neural networks can personalize products by:
a) Analysing large datasets
b) Deleting user data
c) Avoiding recommendations
d) Removing preferences

Q57. Which is a societal concern related to neural networks?
a) Data privacy
b) Faster computation
c) Personalization
d) Innovation

Q58. The simple Celsius-to-Fahrenheit model uses:
a) A single layer with a single neuron
b) Three output layers
c) Two generators
d) Four hidden layers

Q59. In the simple TensorFlow model, units=1 specifies:
a) One dataset
b) One neuron
c) One epoch
d) One feature only

Q60. A loss function measures:
a) How far predictions are from the desired outcome
b) Number of layers
c) Number of features
d) Learning speed only

Q61. The purpose of an optimizer function is to:
a) Increase loss
b) Adjust internal values to reduce loss
c) Add more data
d) Visualize images

Q62. In TensorFlow Playground, one complete iteration through the dataset is called:
a) Batch
b) Epoch
c) Feature
d) Layer

Q63. The learning rate is responsible for:
a) Speed at which the model learns
b) Number of neurons
c) Number of datasets
d) Type of output

Q64. L1 and L2 regularization are used to:
a) Increase overfitting
b) Reduce overfitting
c) Increase the number of features
d) Remove the output layer

Q65. Statement I: Backpropagation adjusts weights based on error.
Statement II: Backpropagation helps improve prediction over time.
a) Only Statement I is true
b) Only Statement II is true
c) Both are true
d) Both are false

Q66. Statement I: A Deep Neural Network has two or more hidden layers.
Statement II: Deep learning refers to the number of hidden layers.
a) Only Statement I is true
b) Only Statement II is true
c) Both are true
d) Both are false

Q67. Statement I: A perceptron is useful for binary classification.
Statement II: A perceptron was created by Frank Rosenblatt in 1958.
a) Only Statement I is true
b) Only Statement II is true
c) Both are true
d) Both are false

Q68. Statement I: In TensorFlow Playground, blue weights represent positive weights.
Statement II: Orange weights represent negative weights.
a) Only Statement I is true
b) Only Statement II is true
c) Both are true
d) Both are false

Q69. A designer wants to generate realistic synthetic images. Which neural network would be most suitable?
a) GAN
b) Perceptron
c) RNN
d) FFNN only

Q70. A neuron receives inputs 2, 3 and 1 with weights 0.4, 0.2 and 0.6 respectively and bias 0.1. What is its weighted sum?
a) 1.8
b) 2.1
c) 2.5
d) 3.0

Q71. In the above situation, if the threshold is 3.0, what will be the output according to the given example?
a) 0
b) 1
c) 2.1
d) 3

Q72. Suppose a neuron has an output of 0.8 and its threshold is 0.5. What will happen according to the threshold logic explained in the chapter?
a) It becomes inactive
b) It becomes active
c) It is removed
d) It changes its weight to zero

Q73. A model’s predictions are consistently far from the desired outcomes. Which component measures this difference?
a) Weight
b) Loss function
c) Input layer
d) Bias

Q74. A student increases the learning rate in TensorFlow Playground and observes changes in how quickly the model learns. Which concept is being investigated?
a) Epoch
b) Learning rate
c) Bias
d) Feature extraction

Q75. A neural network performs extremely well on training data but does not generalize well. Which technique from the chapter can help reduce this issue?
a) Regularization
b) Increasing bias only
c) Removing the output
d) Removing all features

Q76. A student wants to understand how different neural network parameters affect training without dealing heavily with mathematical formulas. Which tool from the chapter would be suitable?
a) TensorFlow Playground
b) CSV editor
c) Python IDLE
d) Spreadsheet

Q77. Assertion (A): CNNs are particularly effective for visual data.
Reason (R): CNNs use filters to extract features from images.
a) Both A and R are true, and R is the correct explanation of A
b) Both A and R are true, but R is not the correct explanation of A
c) A is true, but R is false
d) A is false, but R is true

Q78. Assertion (A): RNNs are suitable for sequential data.
Reason (R): RNNs have feedback connections that allow information persistence across time steps.
a) Both A and R are true, and R is the correct explanation of A
b) Both A and R are true, but R is not the correct explanation of A
c) A is true, but R is false
d) A is false, but R is true

Q79. Assertion (A): A perceptron is useful for binary classification.
Reason (R): A perceptron uses Threshold Logic Units as artificial neurons.
a) Both A and R are true, and R is the correct explanation of A
b) Both A and R are true, but R is not the correct explanation of A
c) A is true, but R is false
d) A is false, but R is true