Advanced AI and Data Science

1. What is "attention mechanism" in neural networks?

• A. Focusing training on one layer
• B. A mechanism allowing models to weigh the importance of different parts of the input when making predictions ✓
• C. A type of regularization
• D. A memory optimization technique

💡 Attention allows models to focus on relevant parts of the input sequence when producing each output, crucial for translation and LLMs.

2. What is "precision" in machine learning metrics?

• A. The ratio of correctly predicted positives to all predicted positives ✓
• B. The ratio of correctly predicted positives to all actual positives
• C. The overall accuracy of a model
• D. The F1 score

💡 Precision = True Positives / (True Positives + False Positives) — measures how many predicted positives are actually positive.

3. What is "overfitting" in machine learning?

• A. A model that is too simple
• B. A model performing well on training data but poorly on unseen data ✓
• C. A model that trains too slowly
• D. A model with too few parameters

💡 Overfitting occurs when a model learns training data too well, including noise, causing poor generalization to new data.

4. What is "federated learning"?

• A. Training on one central server
• B. A distributed ML approach training models locally on devices without sharing raw data ✓
• C. A type of transfer learning
• D. A cloud-based training method

💡 Federated learning trains models across decentralized devices using local data without transferring it centrally, preserving privacy.

5. What is "regularization" in machine learning?

• A. Making a model more complex
• B. Techniques like L1/L2 penalizing model complexity to prevent overfitting ✓
• C. Speeding up training
• D. Normalizing input data

💡 Regularization adds a penalty to the loss function for model complexity (L1/L2), discouraging overfitting by keeping weights small.

6. What is "transfer learning" in AI?

• A. Moving a model between computers
• B. Using a pre-trained model on a new related task with minimal additional training ✓
• C. Training from scratch each time
• D. A type of data augmentation

💡 Transfer learning leverages knowledge from a model trained on one task and adapts it to a different but related task, saving time and data.

7. What is "GAN" in machine learning?

• A. General Adversarial Node
• B. Generative Adversarial Network — two networks (generator and discriminator) competing to create realistic data ✓
• C. Gradient Automated Network
• D. Graph Attention Network

💡 GANs pit a generator (creates fake data) against a discriminator (detects fakes) in a competitive training process to generate realistic content.

8. What is "reinforcement learning"?

• A. Supervised learning with rewards
• B. Learning by trial and error, receiving rewards or penalties based on actions in an environment ✓
• C. Unsupervised learning
• D. A type of deep learning only

💡 Reinforcement learning trains agents to make decisions by maximizing cumulative rewards through interaction with an environment.

9. What does "BERT" stand for in AI?

• A. Binary Encoding Representation Technique
• B. Bidirectional Encoder Representations from Transformers ✓
• C. Basic Encoder Recurrent Transformer
• D. Batch Encoded Recurrent Technique

💡 BERT is a pre-trained language model using bidirectional transformers, trained on masked language modeling to understand context from both directions.

10. What is "backpropagation" in neural networks?

• A. Forward pass through a network
• B. The algorithm computing gradients by propagating errors backward through the network to update weights ✓
• C. A type of activation function
• D. A network initialization method

💡 Backpropagation computes the gradient of the loss function with respect to each weight by applying the chain rule backward through the network.

11. What is "LSTM" in deep learning?

• A. Linear Sequential Training Model
• B. Long Short-Term Memory — a type of RNN designed to learn long-term dependencies ✓
• C. Layered Sequential Transformer Model
• D. Large Scale Training Method

💡 LSTM networks use gating mechanisms (input, forget, output gates) to selectively remember or forget information over long sequences.

12. What is the "vanishing gradient problem"?

• A. Gradients becoming too large
• B. Gradients becoming very small during backpropagation, preventing deep networks from learning effectively ✓
• C. A type of overfitting
• D. A memory issue

💡 In deep networks, gradients can shrink exponentially during backpropagation, making earlier layers learn extremely slowly or not at all.

13. What is "gradient descent" in machine learning?

• A. An optimization algorithm minimizing loss by iteratively adjusting weights in the direction of steepest descent ✓
• B. A type of neural network
• C. A data preprocessing step
• D. A regularization technique

💡 Gradient descent iteratively adjusts model weights by moving in the direction that minimizes the loss function — the core of neural network training.

14. What is "data augmentation" in machine learning?

• A. Deleting redundant data
• B. Artificially increasing training data size by creating modified versions of existing data ✓
• C. Collecting more real data
• D. Normalizing data

💡 Data augmentation creates modified copies of training data (rotations, flips, crops) to increase dataset size and model robustness.

15. What is "A/B testing" in technology?

• A. A type of unit testing
• B. Comparing two versions of a product by exposing different user groups to each version to determine which performs better ✓
• C. A type of security testing
• D. A machine learning evaluation method

💡 A/B testing simultaneously exposes user groups to two variants (A and B) to measure which produces better outcomes based on metrics.

16. What is "confusion matrix" in machine learning evaluation?

• A. A matrix for complex calculations
• B. A table showing predicted vs actual classifications including true/false positives and negatives ✓
• C. A type of neural network layer
• D. A data preprocessing tool

💡 A confusion matrix shows the counts of true positives, true negatives, false positives, and false negatives for a classification model.

17. What is "principal component analysis" (PCA)?

• A. A classification algorithm
• B. A dimensionality reduction technique transforming data to fewer dimensions while preserving variance ✓
• C. A clustering algorithm
• D. A type of neural network

💡 PCA reduces dataset dimensions by finding principal components (directions of maximum variance), removing correlated features.

18. What is a "transformer" architecture in AI?

• A. A type of RNN
• B. An attention-based neural network architecture revolutionizing NLP (used in GPT, BERT) ✓
• C. A type of CNN
• D. A type of reinforcement learning

💡 The Transformer architecture uses self-attention mechanisms to process sequences in parallel, forming the basis of modern LLMs like GPT.

19. What is "hyperparameter tuning"?

• A. Adjusting model weights during training
• B. Optimizing configuration settings (learning rate, layers, batch size) that govern the training process ✓
• C. Normalizing input features
• D. A type of regularization

💡 Hyperparameter tuning finds the optimal configuration settings (not learned from data) that control the training process and model architecture.

20. What is "MapReduce" in big data?

• A. A type of database
• B. A programming model for processing large datasets in parallel across a distributed cluster ✓
• C. A data visualization tool
• D. A type of NoSQL database

💡 MapReduce processes large datasets by splitting tasks into Map (parallel processing) and Reduce (aggregating results) phases across clusters.