AI Intermediate Quiz - Free Practice Quiz

Question 1 of 30

What is backpropagation?

A Running a model in reverse to generate input data B An algorithm that computes gradients of the loss function by propagating errors backward through each network layer C A technique for preprocessing training data D A method for reducing model size

1 / 30

Question 2 of 30

What is gradient descent?

A Moving data down through network layers B A technique for reducing overfitting C A method for initializing weights randomly D An optimization algorithm that iteratively adjusts model weights in the direction that minimizes the loss function

2 / 30

Question 3 of 30

What is a Convolutional Neural Network (CNN) primarily designed for?

A Image recognition and processing tasks B Natural language processing tasks C Time series forecasting D Recommendation systems

3 / 30

Question 4 of 30

What is a Recurrent Neural Network (RNN) designed for?

A Static image classification B Processing tabular data C Sequential or time-series data where order and context matter D Clustering similar data points

4 / 30

Question 5 of 30

What is the vanishing gradient problem?

A Gradients that increase exponentially during training B When gradients shrink exponentially during backpropagation, causing early network layers to learn very slowly C A problem with disappearing training data D When model weights converge to zero

5 / 30

Question 6 of 30

What is the key innovation of the Transformer architecture?

A Using convolutions for sequence modeling B Significantly reducing model parameter count C Processing data without any training D The self-attention mechanism that captures relationships between all positions in a sequence simultaneously

6 / 30

Question 7 of 30

What is transfer learning?

A Reusing a pre-trained model's knowledge as a starting point and adapting it to a new task B Moving a trained model from one computer to another C Transferring labeled data between datasets D Training a model across multiple GPUs

7 / 30

Question 8 of 30

What is regularization used for in machine learning?

A Speeding up model training B Increasing a model's accuracy on training data C Reducing overfitting by penalizing model complexity D Organizing and sorting training data

8 / 30

Question 9 of 30

What is dropout in neural networks?

A Removing training examples from the dataset B Randomly deactivating a fraction of neurons during each training step to prevent overfitting C Reducing the number of layers in a network D Skipping certain training iterations

9 / 30

Question 10 of 30

What does batch normalization do?

A Trains the model in smaller data batches B Normalizes only the final output of the model C Automatically reduces the batch size during training D Normalizes layer inputs within each mini-batch to stabilize activations and accelerate training

10 / 30

Question 11 of 30

What is a loss function?

A A function that measures the difference between a model's predictions and the actual target values B A function that determines which neurons to activate C A function that stops training when accuracy is high enough D A function that randomly initializes model weights

11 / 30

Question 12 of 30

What does the F1 score measure?

A The training speed of a model B The overall accuracy of the model on test data C The harmonic mean of precision and recall D The number of features used by the model

12 / 30

Question 13 of 30

What does a confusion matrix show?

A How confused the model is about its own architecture B A table of true positives, false positives, true negatives, and false negatives broken down by class C The distribution of training examples across classes D The correlation between input features

13 / 30

Question 14 of 30

What does ROC-AUC measure?

A Training speed across different hardware configurations B The number of correctly classified examples C The model's memory usage during inference D The area under the ROC curve — representing how well a model distinguishes between positive and negative classes

14 / 30

Question 15 of 30

What is k-fold cross-validation?

A Splitting data into k subsets and training k models — each using a different subset as the test set — then averaging performance B Running training k times with different hyperparameters C Using k different model architectures and averaging results D Adding k new features to the dataset each iteration

15 / 30

Question 16 of 30

What is feature engineering?

A Designing the architecture of a neural network B Automatically discovering features using deep learning C The process of selecting, transforming, or creating input variables to improve model performance D Removing irrelevant examples from training data

16 / 30

Question 17 of 30

What does PCA (Principal Component Analysis) do?

A Analyzes principal features for classification tasks B Reduces dataset dimensionality by transforming data into uncorrelated principal components ordered by variance C Increases the number of features in the dataset D Clusters data into principal groups

17 / 30

Question 18 of 30

What is a Support Vector Machine (SVM)?

A A type of multi-layer neural network B A hardware component for AI acceleration C A reinforcement learning algorithm D A machine learning algorithm that finds the optimal separating hyperplane that maximizes the margin between classes

18 / 30

Question 19 of 30

What does word embedding (e.g., Word2Vec) do?

A Represents words as dense numerical vectors that capture semantic relationships B Encrypts text data for security purposes C Counts word frequency within documents D Translates words between different languages

19 / 30

Question 20 of 30

What is the attention mechanism in Transformer networks?

A A mechanism that focuses training on the hardest examples B A technique that speeds up convolution operations C A mechanism computing weighted relationships between all elements in a sequence to capture context D A method that automatically selects the most important training data

20 / 30

Question 21 of 30

What is LSTM (Long Short-Term Memory)?

A A CNN variant designed for processing long images B A type of RNN with gating mechanisms (input, forget, output gates) that can learn long-term dependencies C A language model optimized for short text sequences D A memory optimization technique for inference

21 / 30

Question 22 of 30

What is the bias-variance tradeoff?

A The tradeoff between CPU and GPU usage B The tradeoff between training speed and accuracy C The tradeoff between training set size and test set size D The tradeoff between a model's underfitting tendency (high bias) and its sensitivity to training data noise (high variance)

22 / 30

Question 23 of 30

What is data augmentation?

A Artificially increasing training dataset size by creating modified versions of existing examples B Adding more features to existing training data C Removing duplicate records from the dataset D Scaling numerical features to a standard range

23 / 30

Question 24 of 30

What is a Random Forest?

A A dataset collected from forest environments B A neural network with randomly initialized connections C An ensemble method that trains many decision trees on random subsets of data and features, then aggregates predictions — majority vote for classification, averaging for regression D A density-based clustering algorithm

24 / 30

Question 25 of 30

What is precision in the context of classification?

A The overall accuracy of the model across all classes B The fraction of the model's positive predictions that are actually correct C The fraction of actual positives that the model correctly identified D The total number of correct training predictions

25 / 30

Question 26 of 30

What is recall in the context of classification?

A The fraction of positive predictions that are actually correct B The model's ability to memorize training data C The number of examples the model classified correctly D The fraction of actual positive cases that the model successfully identified

26 / 30

Question 27 of 30

What is fine-tuning in the context of Large Language Models?

A Continuing training of a pre-trained model on task-specific or domain-specific data to improve performance on a target task B Adjusting hyperparameters only at inference time C Removing unnecessary layers from a pre-trained model D Optimizing only the model's inference speed

27 / 30

Question 28 of 30

What is XGBoost primarily known for?

A Image classification on visual data B State-of-the-art performance on natural language tasks C Gradient boosted decision trees that deliver high performance on structured and tabular data D Reinforcement learning in complex environments

28 / 30

Question 29 of 30

What is the primary difference between classification and regression?

A Classification uses neural networks; regression uses decision trees B Classification predicts discrete categories; regression predicts continuous numerical values C Regression is always more accurate than classification D Classification always requires significantly more training data

29 / 30

Question 30 of 30

What is prompt engineering?

A Engineering the hardware infrastructure used to run LLMs B Training a language model from scratch using human-crafted examples C Converting natural language prompts into executable code D Designing and optimizing input prompts to guide large language models toward desired outputs without changing model weights

30 / 30