2024-Q1-AI-B 10. Eksāmens

The exam will have 20 randomly selected questions from the given ones, which will need to be answered on paper by circling.Circle one correct answer! (circle the number)

Which statement is correct?
1. Artificial intelligence today is a complex computer program that mainly consists of programming rules
2. Artificial intelligence today is a mathematical model that mainly consists of mathematical equations
3. Artificial intelligence today is a complex computer program that mainly consists of programmed expert knowledge
What does artificial intelligence learn from?
1. Rules created by an expert
2. Data
3. Rules created by a programmer
Which of the given examples could be input data in an artificial intelligence model?
1. Probability that a customer will stop using a service
2. How many times a customer has logged into the system in the last 10 days
3. Model weight values
Which of the given examples could be output data in an artificial intelligence model?
1. Probability that a customer will stop using a service
2. How many times a customer has logged into the system in the last 10 days
3. Model weight values
To predict a product price, what type of model is needed?
1. Regression
2. Classification
3. Numeration
To predict whether a customer will stop using a service, what type of model is needed?
1. Regression
2. Classification
3. Numeration
In which environment is Artificial Intelligence usually trained?
1. Excel
2. Python
3. Power BI
What data sets are needed to train a model that could be used in production?
1. Training set
2. Test set
3. Validation set
4. Training, Test, Validation sets
Which factor most influences model accuracy?
1. Learning rate
2. Unbalanced number of samples in each class in the training data set
3. Diversity of samples in the data set
For what application would artificial intelligence not be effective?
1. Writing text ads
2. Checking passwords and usernames when logging into websites
3. Making coloring books for children
4. Composing music
How similar is the artificial deep neural network model to the human natural neural network model?
1. Almost identical, as evidenced by large language models, image models and other models
2. Very similar because it models biochemical processes
3. Not similar, because the artificial neural network model is mathematical and works very differently from the human natural neural network
Which sequence of actions corresponds to the training of deep neural network models?
1. Data normalization, Data splitting, Model creation, Error function selection, Additional metric selection, Test cycle, Validation cycle, Epochs, Training cycle, Backpropagation
2. Data normalization, Data splitting, Model creation, Epochs, Training cycle, Backpropagation, Error function selection, Additional metric selection, Test cycle, Validation cycle
3. Data normalization, Data splitting, Model creation, Error function selection, Additional metric selection, Epochs, Training cycle, Backpropagation, Test cycle, Validation cycle
What does Epoch mean in the training process of artificial neural networks?
1. All samples in the training set are considered and there can be many Epochs in one training process
2. Data normalization method that removes extreme values
3. All samples in the training set are considered and there can be only one Epoch in the training process
4. Validation samples are considered after training
If the numerical value of the MSE error function is 0.5, then after one training step, the numerical value is most likely: 1. 0.6 2. 0.5 3. 0.4
RNN is usually used to: 1. Recognize multiple objects in an image 2. Predict stock prices from stock market data 3. Predict car prices from an advertisement
ConvNet without data augmentation during training is able to recognize: 1. Objects moved in the image 2. Objects moved and rotated in the image 3. Objects moved, enlarged and rotated in the image
In a pre-trained RNN, the weights W at each time step: 1. are different 2. are the same 3. is not defined
The Transformer model is based on:
1. Memory variable in the hidden vector h_t
2. Attention mechanism
3. Both
Which component is the most important in ChatGPT query engineering to achieve a quality response?
1. Formulating the query in the shortest and most precise form possible
2. Formulating the query in the longest and most extensive form possible
3. Copying facts into the query
What will happen if you continue to ask multiple questions on different topics in the same ChatGPT session?
1. The language model will start copying content from previous questions in subsequent answers
2. It does not affect the language model's operation
3. The language model will get confused and not know what to answer
In a pre-trained GRU, the weights W at each time step:
1. are different
2. are the same
3. is not defined

What are A2C, PPO models used for (Reinforcement Learning)?
1. To train the sensor fusion of a spacecraft
2. To train ChatGPT to respond with responses that people usually expect in chat
3. To train to predict stock prices

How does ViT (Vision Transformer) differ from GPT-type Transformers?
1. Identical, no differences
2. The main difference between ViT and GPT is that in the initial layers, the image is divided into regions, then a Linear Layer is passed to the Transformer
3. The main difference between ViT and GPT is that there is no transposed word embedding matrix at the end
How do UNet and YOLO models differ?
1. UNet performs semantic segmentation, while YOLO performs object recognition
2. UNet performs semantic segmentation and is able to recognize each object separately, while YOLO performs object recognition, where all objects from one class are recognized together
3. UNet performs instance segmentation, while YOLO performs object recognition
How does DenseNet differ from ResNet?
1. ResNet has one skip connection for each block using addition, but DenseNet has one skip connection using multiplication
2. ResNet has one skip connection for each block using multiplication, but DenseNet has multiple skip connections connecting multiple layers forward using multiplication
3. ResNet has one skip connection for each block using addition, but DenseNet has multiple skip connections connecting multiple layers forward using addition
What does DQN, DDQN model?
1. State value function
2. Q-Policy function that returns probabilities of actions
3. Predicted cumulative reward function for state and action trajectory
How to encode Time embeddings in the Transformer model?
1. Using a cosine-sine hard-coded table with embeddings
2. Using a trainable embedding table
3. Both ways

What is the matrix dot product (dot product)?
1. A mathematical operation that obtains a perpendicular vector or matrix between input vectors
2. A mathematical operation that performs matrix transformation using multiplication in any dimension
3. An algorithm that uses addition and multiplication in the last 2 dimensions of any matrix
What is a Linear layer or function in artificial neural networks?
1. Matrix vectorial product
2. Matrix dot product and bias with addition
3. Linear regression algorithm
Why is batch normalization needed before the activation function?
1. To prevent overfitting
2. To prevent dead neurons
3. To prevent bias towards one class in predictions
Why is an activation function needed in artificial neural networks?
1. To provide nonlinearity
2. To provide linearity
3. To provide balance
How to enter categorical values in an artificial neural network?
1. As categorical values
2. As One-Hot-Encoded vectors
3. As category indexes
4. As category indexes, but with normalization
Why is it necessary to divide the training data sets for training, testing, and validation?
1. To prevent overfitting
2. To prevent dead neurons
3. To prevent bias towards one class in predictions
How to format categorical output data in an artificial neural network?
1. As trainable embeddings
2. As One-Hot-Encoded vectors
3. As category indexes
What Q-Values can be predicted with Deep Q-Learning Network (DQN)?
1. Discrete
2. Continuous (stochastic)
3. Both
What error function is used to train artificial neural networks for classification?
1. Mean Squared Error (MSE)
2. Mean Absolute Error (MAE)
3. Categorical Cross-Entropy
In which cases should the MAE error function be used?
1. When the training data is clean and not noisy
2. When the training data is noisy
3. When the training data has smaller size dataset
With what models can time series tasks be solved?
1. LSTM
2. GRU
3. Transformer
4. All of the above
What is needed to train a Reinforcement Learning model?
1. Training data
2. Training environment and reward values
3. Training data and environment

How to achieve a quality response from ChatGPT?
1. Give the task, format, facts, and examples
2. Formulate the query in the shortest and most precise form possible
3. Ask as many questions as possible in the same session
Can multiple questions on different topics be asked in the same ChatGPT session?
1. Yes
2. No
3. Only if the topics are related

How to achieve a quality response from ChatGPT?
1. Give the task in multiple parts as precisely as possible in one query
2. Give one task at a time in each query
3. It doesn't matter how many steps are in one task execution
What can be achieved with a matrix scalar product (dot product)?
1. Reduce the dimensions of the hidden vector
2. Increase the dimensions of the hidden vector
3. Both mentioned
Why can't an artificial neural network be composed only of linear layers?
1. The result will be too complex
2. The result will be too simple
3. The result will be reduced to the last linear function, ignoring the previous ones
4. It will not be possible to perform the backpropagation algorithm
What does the backpropagation algorithm consist of?
1. SGD (stochastic gradient descent)
2. Calculation of the derivative of the error function by weights and SGD (stochastic gradient descent)
3. Calculation of the derivative of the weights by the error function and SGD (stochastic gradient descent)
What are embeddings?
1. Hidden values obtained from numerical values
2. Categorical values converted to numerical values using One-Hot-Encoding
3. Categorical values converted to numerical values using trainable parameters

What are Adam, RMSProp, SGD algorithms?
1. Error function stabilization algorithm
2. Normalization algorithm
3. Optimization algorithm
How can you train a model on a GPU?
1. Using the Matplotlib library
2. Using the CUDA libraries that are built into TensorFlow, PyTorch
3. Using the NumPy library
What augmentations are needed for image classification tasks?
1. Image translation
2. Image rotation
3. All mentioned
How many dimensions should the output layer have to classify 10 categories?
1. 1
2. 10
3. 2
4. Not specified
How many dimensions should the output layer have to classify 2 categories?
1. 1
2. 2
3. Not specified

What function should follow before the Cross-Entropy function to classify 10 categories?
1. Softmax
2. Sigmoid
3. ReLU

What function should follow before the Binary-Cross-Entropy function to classify 2 categories?
1. Softmax
2. Sigmoid
3. ReLU
What limitations are for ChatGPT?
1. Context length
2. Limited training data
3. Both above
What should be done to input data in tabular datasets?
1. Feature scaling (normalization)
2. Feature extraction
3. Both above
What does PCA algorithm do?
1. Reduces the number of features, dimensionality reduction
2. Increases the number of features, dimensionality increase
3. Both above
What does XBoost, ID3 algorithms do?
1. Neural network algorithms
2. Decision tree algorithms
3. Recurrent neural network algorithms
What does K-means, DBSCAN algorithms do?
1. Clustering algorithms
2. Classification algorithms
3. Regression algorithms
What does the Random Forest algorithm do?
1. Ensemble learning algorithm
2. Clustering algorithm
3. Both above
Is it possible to predict for car advertisement price and number of owners using the same model at once?
1. Yes
2. No
3. Only if values are correlated
What is the main hyperparameter in the spectral clustering algorithm?
1. Number of clusters
2. Number of samples
3. Number of feature
When to use Decision Tree algorithm instead of Deep Neural Network?
1. When the dataset is noisy
2. When the dataset is clean
3. When the dataset is small