Artificial Intelligence Cheat Sheet 2026

The 30 highest-yield Artificial Intelligence facts, distilled from real exam questions. Print it, save it as a PDF, or study it here — free, no sign-up.

1. What does 'dropout' do during neural network training? → Randomly deactivates neurons during each training step to reduce overfitting
2. Which architecture introduced residual (skip) connections to train very deep networks? → ResNet
3. What distinguishes a convolutional neural network (CNN) from a standard feedforward network? → CNNs apply learnable filters that share weights spatially
4. Which architecture introduced the concept of encoder-decoder with skip connections, widely used for image segmentation? → U-Net
5. What is 'human-in-the-loop' AI? → A system where human judgment is incorporated at critical decision points
6. Which of the following describes narrow AI? → AI specialized in a single task like playing chess
7. What is the purpose of a pooling layer in a CNN? → Reduce spatial dimensions and provide translational invariance
8. What is 'image classification' in computer vision? → Assigning a single label to an entire image based on its content
9. Which evaluation approach splits data into multiple folds, training and testing on different combinations? → Cross-validation
10. Which of the following is a primary goal of Artificial Intelligence research? → Reasoning
11. What does 'epoch' mean in neural network training? → One complete pass through the entire training dataset
12. What is a 'frame' in the context of knowledge representation? → A data structure that groups an object's attributes and default values into a single unit
13. Which of the following search algorithms uses the least amount of memory? → Depth First Search
14. What is 'explainability' (or interpretability) in AI? → The degree to which humans can understand and trace how an AI makes decisions
15. Which scenario best illustrates algorithmic bias in AI? → A hiring model penalizing resumes from certain demographics due to skewed training data
16. What is 'consent' in AI data collection? → Individuals knowingly and voluntarily agreeing to have their data collected and used
17. What does 'backpropagation' compute in a neural network training cycle? → The gradient of the loss with respect to each network weight
18. What does regularization in machine learning primarily address? → Overfitting
19. What is the purpose of a validation set in machine learning? → To tune hyperparameters and detect overfitting
20. The role of an inference engine in an expert system is to: → Apply logical rules to the knowledge base to derive new facts or conclusions
21. What is backpropagation in neural network training? → Using the chain rule to compute gradients and update weights
22. What does 'fine-tuning' a pre-trained language model involve? → Continuing training on a task-specific labeled dataset to adapt the model to a new task
23. Which of the following is an Artificial Intelligence tool? → Neural networks
24. What does backpropagation compute during neural network training? → Gradients of the loss with respect to each weight
25. Which of these is a closely related field to AI? → Mathematics
26. Which query language is used to retrieve and manipulate data stored in RDF (Resource Description Framework) knowledge graphs? → SPARQL
27. In machine learning, what problem occurs when a model performs well on training data but poorly on unseen data? → Overfitting
28. What problem do LSTMs (Long Short-Term Memory networks) solve compared to standard RNNs? → Vanishing gradients over long sequences
29. What is a confusion matrix used to evaluate? → Classification model performance across all classes
30. Which word embedding model learns vector representations by predicting surrounding words (skip-gram) or predicting a word from context (CBOW)? → Word2Vec