AI Deep Learning Basics MCQ Questions with Answers (Latest 2026)

Q1. Which option best describes a neuron?

A. A model loss value. B. A unit computing a weighted sum followed by a nonlinearity. C. A token embedding only. D. A retrieval shard.

Select an answer to check.

Answer: A unit computing a weighted sum followed by a nonlinearity.

Here, A unit computing a weighted sum followed by a nonlinearity. is the right choice. Building block of neural nets. It aligns directly with what the question asks about which option best describes a neuron. A quick elimination of partially true options helps confirm it.

Q2. What is the primary purpose of a neuron?

A. A retrieval shard. B. A token embedding only. C. A unit computing a weighted sum followed by a nonlinearity. D. A model loss value.

Select an answer to check.

Answer: A unit computing a weighted sum followed by a nonlinearity.

In this case, A unit computing a weighted sum followed by a nonlinearity. is correct. Building block of neural nets. It aligns directly with what the question asks about what is the primary purpose of a neuron. A quick elimination of partially true options helps confirm it.

Q3. Which statement about a neuron is most accurate?

A. A token embedding only. B. A model loss value. C. A retrieval shard. D. A unit computing a weighted sum followed by a nonlinearity.

Select an answer to check.

Answer: A unit computing a weighted sum followed by a nonlinearity.

The best option here is A unit computing a weighted sum followed by a nonlinearity.. Building block of neural nets. It aligns directly with what the question asks about which statement about a neuron is most accurate. A quick elimination of partially true options helps confirm it.

Q4. How is a neuron best characterized?

A. A retrieval shard. B. A model loss value. C. A unit computing a weighted sum followed by a nonlinearity. D. A token embedding only.

Select an answer to check.

Answer: A unit computing a weighted sum followed by a nonlinearity.

For this question, A unit computing a weighted sum followed by a nonlinearity. is correct. Building block of neural nets. It aligns directly with what the question asks about how is a neuron best characterized. A quick elimination of partially true options helps confirm it.

Q5. Which option best describes an activation function?

A. Nonlinearity applied to a neuron's pre-activation. B. A regularization term. C. A loss function. D. An optimizer step.

Select an answer to check.

Answer: Nonlinearity applied to a neuron's pre-activation.

Nonlinearity applied to a neuron's pre-activation. is the correct answer here. Examples: ReLU, GELU, sigmoid. It aligns directly with what the question asks about which option best describes an activation function. A quick elimination of partially true options helps confirm it.

Q6. What is the primary purpose of an activation function?

A. An optimizer step. B. Nonlinearity applied to a neuron's pre-activation. C. A regularization term. D. A loss function.

Select an answer to check.

Answer: Nonlinearity applied to a neuron's pre-activation.

Here, Nonlinearity applied to a neuron's pre-activation. is the right choice. Examples: ReLU, GELU, sigmoid. This matches the core idea being tested around what is the primary purpose of an activation. A quick elimination of partially true options helps confirm it.

Q7. Which statement about an activation function is most accurate?

A. An optimizer step. B. Nonlinearity applied to a neuron's pre-activation. C. A regularization term. D. A loss function.

Select an answer to check.

Answer: Nonlinearity applied to a neuron's pre-activation.

In this case, Nonlinearity applied to a neuron's pre-activation. is correct. Examples: ReLU, GELU, sigmoid. This matches the core idea being tested around which statement about an activation function is most. A quick elimination of partially true options helps confirm it.

Q8. How is an activation function best characterized?

A. An optimizer step. B. A regularization term. C. Nonlinearity applied to a neuron's pre-activation. D. A loss function.

Select an answer to check.

Answer: Nonlinearity applied to a neuron's pre-activation.

The best option here is Nonlinearity applied to a neuron's pre-activation.. Examples: ReLU, GELU, sigmoid. This matches the core idea being tested around how is an activation function best characterized. A quick elimination of partially true options helps confirm it.

Q9. Which option best describes ReLU?

A. A sigmoid squashing to (0,1). B. A loss function. C. max(0, x); piecewise linear activation. D. A softmax over classes.

Select an answer to check.

Answer: max(0, x); piecewise linear activation.

For this question, max(0, x); piecewise linear activation. is correct. Cheap and effective default. This matches the core idea being tested around which option best describes relu. A quick elimination of partially true options helps confirm it.

Q10. What is the primary purpose of ReLU?

A. A softmax over classes. B. max(0, x); piecewise linear activation. C. A loss function. D. A sigmoid squashing to (0,1).

Select an answer to check.

Answer: max(0, x); piecewise linear activation.

max(0, x); piecewise linear activation. is the correct answer here. Cheap and effective default. This matches the core idea being tested around what is the primary purpose of relu. A quick elimination of partially true options helps confirm it.

Q11. Which statement about ReLU is most accurate?

A. A sigmoid squashing to (0,1). B. max(0, x); piecewise linear activation. C. A softmax over classes. D. A loss function.

Select an answer to check.

Answer: max(0, x); piecewise linear activation.

Here, max(0, x); piecewise linear activation. is the right choice. Cheap and effective default. That is exactly the concept behind which statement about relu is most accurate in this context. A quick elimination of partially true options helps confirm it.

Q12. How is ReLU best characterized?

A. A loss function. B. A sigmoid squashing to (0,1). C. A softmax over classes. D. max(0, x); piecewise linear activation.

Select an answer to check.

Answer: max(0, x); piecewise linear activation.

In this case, max(0, x); piecewise linear activation. is correct. Cheap and effective default. That is exactly the concept behind how is relu best characterized in this context. A quick elimination of partially true options helps confirm it.

Q13. Which option best describes softmax?

A. An identity function. B. A binary thresholding operator. C. Converts logits to a probability distribution over classes. D. An L2 regularizer.

Select an answer to check.

Answer: Converts logits to a probability distribution over classes.

The best option here is Converts logits to a probability distribution over classes.. Used in multi-class classification heads. That is exactly the concept behind which option best describes softmax in this context. A quick elimination of partially true options helps confirm it.

Q14. What is the primary purpose of softmax?

A. An L2 regularizer. B. Converts logits to a probability distribution over classes. C. An identity function. D. A binary thresholding operator.

Select an answer to check.

Answer: Converts logits to a probability distribution over classes.

For this question, Converts logits to a probability distribution over classes. is correct. Used in multi-class classification heads. That is exactly the concept behind what is the primary purpose of softmax in this context. A quick elimination of partially true options helps confirm it.

Q15. Which statement about softmax is most accurate?

A. An L2 regularizer. B. A binary thresholding operator. C. Converts logits to a probability distribution over classes. D. An identity function.

Select an answer to check.

Answer: Converts logits to a probability distribution over classes.

Converts logits to a probability distribution over classes. is the correct answer here. Used in multi-class classification heads. That is exactly the concept behind which statement about softmax is most accurate in this context. A quick elimination of partially true options helps confirm it.

Q16. How is softmax best characterized?

A. An identity function. B. Converts logits to a probability distribution over classes. C. A binary thresholding operator. D. An L2 regularizer.

Select an answer to check.

Answer: Converts logits to a probability distribution over classes.

Here, Converts logits to a probability distribution over classes. is the right choice. Used in multi-class classification heads. It fits the requirement in the prompt about how is softmax best characterized. A quick elimination of partially true options helps confirm it.

Q17. Which option best describes backpropagation?

A. A retrieval method. B. A regularization scheme. C. A type of optimizer. D. Algorithm to compute gradients via chain rule through the graph.

Select an answer to check.

Answer: Algorithm to compute gradients via chain rule through the graph.

In this case, Algorithm to compute gradients via chain rule through the graph. is correct. Enables training deep nets. It fits the requirement in the prompt about which option best describes backpropagation. A quick elimination of partially true options helps confirm it.

Q18. What is the primary purpose of backpropagation?

A. Algorithm to compute gradients via chain rule through the graph. B. A retrieval method. C. A regularization scheme. D. A type of optimizer.

Select an answer to check.

Answer: Algorithm to compute gradients via chain rule through the graph.

The best option here is Algorithm to compute gradients via chain rule through the graph.. Enables training deep nets. It fits the requirement in the prompt about what is the primary purpose of backpropagation. A quick elimination of partially true options helps confirm it.

Q19. Which statement about backpropagation is most accurate?

A. Algorithm to compute gradients via chain rule through the graph. B. A retrieval method. C. A type of optimizer. D. A regularization scheme.

Select an answer to check.

Answer: Algorithm to compute gradients via chain rule through the graph.

For this question, Algorithm to compute gradients via chain rule through the graph. is correct. Enables training deep nets. It fits the requirement in the prompt about which statement about backpropagation is most accurate. A quick elimination of partially true options helps confirm it.

Q20. How is backpropagation best characterized?

A. A regularization scheme. B. A retrieval method. C. Algorithm to compute gradients via chain rule through the graph. D. A type of optimizer.

Select an answer to check.

Answer: Algorithm to compute gradients via chain rule through the graph.

Algorithm to compute gradients via chain rule through the graph. is the correct answer here. Enables training deep nets. It fits the requirement in the prompt about how is backpropagation best characterized. A quick elimination of partially true options helps confirm it.

Q21. Which option best describes a feedforward network?

A. A recurrent network with cycles. B. A diffusion model. C. Layers connected sequentially with no cycles. D. A graph neural network.

Select an answer to check.

Answer: Layers connected sequentially with no cycles.

Here, Layers connected sequentially with no cycles. is the right choice. Most basic deep architecture. This is the most accurate statement for which option best describes a feedforward network. A quick elimination of partially true options helps confirm it.

Q22. What is the primary purpose of a feedforward network?

A. A graph neural network. B. A diffusion model. C. A recurrent network with cycles. D. Layers connected sequentially with no cycles.

Select an answer to check.

Answer: Layers connected sequentially with no cycles.

In this case, Layers connected sequentially with no cycles. is correct. Most basic deep architecture. This is the most accurate statement for what is the primary purpose of a feedforward. A quick elimination of partially true options helps confirm it.

Q23. Which statement about a feedforward network is most accurate?

A. A recurrent network with cycles. B. Layers connected sequentially with no cycles. C. A graph neural network. D. A diffusion model.

Select an answer to check.

Answer: Layers connected sequentially with no cycles.

The best option here is Layers connected sequentially with no cycles.. Most basic deep architecture. This is the most accurate statement for which statement about a feedforward network is most. A quick elimination of partially true options helps confirm it.

Q24. How is a feedforward network best characterized?

A. A graph neural network. B. A diffusion model. C. A recurrent network with cycles. D. Layers connected sequentially with no cycles.

Select an answer to check.

Answer: Layers connected sequentially with no cycles.

For this question, Layers connected sequentially with no cycles. is correct. Most basic deep architecture. This is the most accurate statement for how is a feedforward network best characterized. A quick elimination of partially true options helps confirm it.

Q25. Which option best describes a CNN?

A. A purely fully-connected network. B. A pure attention model. C. A retrieval-only model. D. Network using convolutional layers for spatial data.

Select an answer to check.

Answer: Network using convolutional layers for spatial data.

Network using convolutional layers for spatial data. is the correct answer here. Excels on images. This is the most accurate statement for which option best describes a cnn. A quick elimination of partially true options helps confirm it.

Q26. What is the primary purpose of a CNN?

A. Network using convolutional layers for spatial data. B. A pure attention model. C. A purely fully-connected network. D. A retrieval-only model.

Select an answer to check.

Answer: Network using convolutional layers for spatial data.

Here, Network using convolutional layers for spatial data. is the right choice. Excels on images. It aligns directly with what the question asks about what is the primary purpose of a cnn. The other options are either incomplete or contextually incorrect.

Q27. Which statement about a CNN is most accurate?

A. A purely fully-connected network. B. Network using convolutional layers for spatial data. C. A retrieval-only model. D. A pure attention model.

Select an answer to check.

Answer: Network using convolutional layers for spatial data.

In this case, Network using convolutional layers for spatial data. is correct. Excels on images. It aligns directly with what the question asks about which statement about a cnn is most accurate. The other options are either incomplete or contextually incorrect.

Q28. How is a CNN best characterized?

A. Network using convolutional layers for spatial data. B. A retrieval-only model. C. A purely fully-connected network. D. A pure attention model.

Select an answer to check.

Answer: Network using convolutional layers for spatial data.

The best option here is Network using convolutional layers for spatial data.. Excels on images. It aligns directly with what the question asks about how is a cnn best characterized. The other options are either incomplete or contextually incorrect.

Q29. Which option best describes an RNN?

A. A purely feedforward network. B. A pure CNN. C. A static lookup table. D. Network processing sequences via recurrent state.

Select an answer to check.

Answer: Network processing sequences via recurrent state.

For this question, Network processing sequences via recurrent state. is correct. Earlier sequence model paradigm. It aligns directly with what the question asks about which option best describes an rnn. The other options are either incomplete or contextually incorrect.

Q30. What is the primary purpose of an RNN?

A. Network processing sequences via recurrent state. B. A pure CNN. C. A static lookup table. D. A purely feedforward network.

Select an answer to check.

Answer: Network processing sequences via recurrent state.

Network processing sequences via recurrent state. is the correct answer here. Earlier sequence model paradigm. It aligns directly with what the question asks about what is the primary purpose of an rnn. The other options are either incomplete or contextually incorrect.

Q31. Which statement about an RNN is most accurate?

A. A static lookup table. B. Network processing sequences via recurrent state. C. A purely feedforward network. D. A pure CNN.

Select an answer to check.

Answer: Network processing sequences via recurrent state.

Here, Network processing sequences via recurrent state. is the right choice. Earlier sequence model paradigm. This matches the core idea being tested around which statement about an rnn is most accurate. The other options are either incomplete or contextually incorrect.

Q32. How is an RNN best characterized?

A. A static lookup table. B. Network processing sequences via recurrent state. C. A purely feedforward network. D. A pure CNN.

Select an answer to check.

Answer: Network processing sequences via recurrent state.

In this case, Network processing sequences via recurrent state. is correct. Earlier sequence model paradigm. This matches the core idea being tested around how is an rnn best characterized. The other options are either incomplete or contextually incorrect.

Q33. Which option best describes an LSTM?

A. A linear classifier. B. RNN variant with gating to mitigate vanishing gradients. C. A vanilla RNN. D. A purely fully-connected network.

Select an answer to check.

Answer: RNN variant with gating to mitigate vanishing gradients.

The best option here is RNN variant with gating to mitigate vanishing gradients.. Long short-term memory. This matches the core idea being tested around which option best describes an lstm. The other options are either incomplete or contextually incorrect.

Q34. What is the primary purpose of an LSTM?

A. A vanilla RNN. B. RNN variant with gating to mitigate vanishing gradients. C. A linear classifier. D. A purely fully-connected network.

Select an answer to check.

Answer: RNN variant with gating to mitigate vanishing gradients.

For this question, RNN variant with gating to mitigate vanishing gradients. is correct. Long short-term memory. This matches the core idea being tested around what is the primary purpose of an lstm. The other options are either incomplete or contextually incorrect.

Q35. Which statement about an LSTM is most accurate?

A. A vanilla RNN. B. A purely fully-connected network. C. RNN variant with gating to mitigate vanishing gradients. D. A linear classifier.

Select an answer to check.

Answer: RNN variant with gating to mitigate vanishing gradients.

RNN variant with gating to mitigate vanishing gradients. is the correct answer here. Long short-term memory. This matches the core idea being tested around which statement about an lstm is most accurate. The other options are either incomplete or contextually incorrect.

Q36. How is an LSTM best characterized?

A. A linear classifier. B. A vanilla RNN. C. RNN variant with gating to mitigate vanishing gradients. D. A purely fully-connected network.

Select an answer to check.

Answer: RNN variant with gating to mitigate vanishing gradients.

Here, RNN variant with gating to mitigate vanishing gradients. is the right choice. Long short-term memory. That is exactly the concept behind how is an lstm best characterized in this context. The other options are either incomplete or contextually incorrect.

Q37. Which option best describes a Transformer?

A. A linear regression model. B. A pure RNN. C. A pure CNN. D. Architecture based on self-attention.

Select an answer to check.

Answer: Architecture based on self-attention.

In this case, Architecture based on self-attention. is correct. Backbone of modern LLMs. That is exactly the concept behind which option best describes a transformer in this context. The other options are either incomplete or contextually incorrect.

Q38. What is the primary purpose of a Transformer?

A. Architecture based on self-attention. B. A pure RNN. C. A linear regression model. D. A pure CNN.

Select an answer to check.

Answer: Architecture based on self-attention.

The best option here is Architecture based on self-attention.. Backbone of modern LLMs. That is exactly the concept behind what is the primary purpose of a transformer in this context. The other options are either incomplete or contextually incorrect.

Q39. Which statement about a Transformer is most accurate?

A. Architecture based on self-attention. B. A pure CNN. C. A linear regression model. D. A pure RNN.

Select an answer to check.

Answer: Architecture based on self-attention.

For this question, Architecture based on self-attention. is correct. Backbone of modern LLMs. That is exactly the concept behind which statement about a transformer is most accurate in this context. The other options are either incomplete or contextually incorrect.

Q40. How is a Transformer best characterized?

A. Architecture based on self-attention. B. A pure CNN. C. A pure RNN. D. A linear regression model.

Select an answer to check.

Answer: Architecture based on self-attention.

Architecture based on self-attention. is the correct answer here. Backbone of modern LLMs. That is exactly the concept behind how is a transformer best characterized in this context. The other options are either incomplete or contextually incorrect.

Q41. Which option best describes self-attention?

A. A pure pooling layer. B. A token dropout. C. A 1D convolution. D. Each token attends to others to mix context.

Select an answer to check.

Answer: Each token attends to others to mix context.

Here, Each token attends to others to mix context. is the right choice. Key mechanism in Transformers. It fits the requirement in the prompt about which option best describes self-attention. The other options are either incomplete or contextually incorrect.

Q42. What is the primary purpose of self-attention?

A. A token dropout. B. A 1D convolution. C. Each token attends to others to mix context. D. A pure pooling layer.

Select an answer to check.

Answer: Each token attends to others to mix context.

In this case, Each token attends to others to mix context. is correct. Key mechanism in Transformers. It fits the requirement in the prompt about what is the primary purpose of self-attention. The other options are either incomplete or contextually incorrect.

Q43. Which statement about self-attention is most accurate?

A. A 1D convolution. B. A token dropout. C. Each token attends to others to mix context. D. A pure pooling layer.

Select an answer to check.

Answer: Each token attends to others to mix context.

The best option here is Each token attends to others to mix context.. Key mechanism in Transformers. It fits the requirement in the prompt about which statement about self-attention is most accurate. The other options are either incomplete or contextually incorrect.

Q44. How is self-attention best characterized?

A. Each token attends to others to mix context. B. A token dropout. C. A 1D convolution. D. A pure pooling layer.

Select an answer to check.

Answer: Each token attends to others to mix context.

For this question, Each token attends to others to mix context. is correct. Key mechanism in Transformers. It fits the requirement in the prompt about how is self-attention best characterized. The other options are either incomplete or contextually incorrect.

Q45. Which option best describes multi-head attention?

A. Multiple attention heads run in parallel. B. A pooling operation only. C. A single global softmax only. D. Sequential attention only.

Select an answer to check.

Answer: Multiple attention heads run in parallel.

Multiple attention heads run in parallel. is the correct answer here. Captures diverse relationships. It fits the requirement in the prompt about which option best describes multi-head attention. The other options are either incomplete or contextually incorrect.

Q46. What is the primary purpose of multi-head attention?

A. A pooling operation only. B. Multiple attention heads run in parallel. C. Sequential attention only. D. A single global softmax only.

Select an answer to check.

Answer: Multiple attention heads run in parallel.

Here, Multiple attention heads run in parallel. is the right choice. Captures diverse relationships. This is the most accurate statement for what is the primary purpose of multi-head attention. The other options are either incomplete or contextually incorrect.

Q47. Which statement about multi-head attention is most accurate?

A. Multiple attention heads run in parallel. B. A pooling operation only. C. A single global softmax only. D. Sequential attention only.

Select an answer to check.

Answer: Multiple attention heads run in parallel.

In this case, Multiple attention heads run in parallel. is correct. Captures diverse relationships. This is the most accurate statement for which statement about multi-head attention is most accurate. The other options are either incomplete or contextually incorrect.

Q48. How is multi-head attention best characterized?

A. Sequential attention only. B. Multiple attention heads run in parallel. C. A pooling operation only. D. A single global softmax only.

Select an answer to check.

Answer: Multiple attention heads run in parallel.

The best option here is Multiple attention heads run in parallel.. Captures diverse relationships. This is the most accurate statement for how is multi-head attention best characterized. The other options are either incomplete or contextually incorrect.

Q49. Which option best describes positional encoding?

A. A regularizer. B. Encodes order info for attention models. C. A loss function. D. A normalization layer.

Select an answer to check.

Answer: Encodes order info for attention models.

For this question, Encodes order info for attention models. is correct. Required since attention is permutation-invariant. This is the most accurate statement for which option best describes positional encoding. The other options are either incomplete or contextually incorrect.

Q50. What is the primary purpose of positional encoding?

A. Encodes order info for attention models. B. A normalization layer. C. A regularizer. D. A loss function.

Select an answer to check.

Answer: Encodes order info for attention models.

Encodes order info for attention models. is the correct answer here. Required since attention is permutation-invariant. This is the most accurate statement for what is the primary purpose of positional encoding. The other options are either incomplete or contextually incorrect.