TP Test Practice

Upload rules

Prepare your test before uploading

Spend one minute with the guide so questions, answer choices, correct answers and images import cleanly without extra manual fixes.

Read the guide

Test set

Инно ИС для подготовки

Programming · Инно ИС для подготовки.txt

Export

Flashcards

Инно ИС для подготовки

Choose a study flow.

Sequential mode

In order

No repeats

Mastery mode

Until correct

Repeat wrong
150questions
0review
0no answer
3attempts

Discussion

Comments, notes and shared explanations will appear here.

Teacher noteAttach hints to confusing questions.
Learner threadAsk why an answer is correct.
#1

Which statement correctly describes the difference between classification and regression?

  1. Classification predicts continuous values, while regression predicts categories
  2. Classification predicts categories, while regression predicts continuous values
  3. Both classification and regression always predict only binary labels
  4. Regression is used only for unsupervised learning
#2

Which of the following is an example of a regression task?

  1. Detecting whether an email is spam or not
  2. Classifying a tumor as benign or malignant
  3. Predicting the price of a house
  4. Grouping customers into clusters
#3

Which of the following is an example of a classification task?

  1. Predicting tomorrow's temperature
  2. Predicting whether a patient has a disease or not
  3. Estimating the price of a car
  4. Reducing the number of features in a dataset
#4

What is the main characteristic of supervised machine learning?

  1. The model learns without any target labels
  2. The model learns from input data and known target labels
  3. The model only visualizes data
  4. The model randomly assigns clusters
#5

What is the main characteristic of unsupervised machine learning?

  1. It requires labeled training examples
  2. It always predicts numerical target values
  3. It finds patterns in data without target labels
  4. It can only be used for classification
#6

What is the purpose of training data in machine learning?

  1. To teach the model patterns from examples
  2. To test the model only after deployment
  3. To remove all features from the dataset
  4. To manually write all prediction rules
#7

Why is test data used in machine learning?

  1. To train the model parameters
  2. To evaluate how well the model performs on unseen data
  3. To increase the number of training examples
  4. To replace feature scaling
#8

What does generalization mean in machine learning?

  1. Memorizing all training examples perfectly
  2. Performing well on new, unseen data
  3. Using only one feature for prediction
  4. Removing the test set from the workflow
#9

What is overfitting?

  1. A model is too simple to learn patterns
  2. A model fits training data too closely and performs worse on new data
  3. A model has no parameters
  4. A model works only with unlabeled data
#10

What is underfitting?

  1. A model is too simple and fails to capture important patterns
  2. A model learns training data too well
  3. A model has too many decision rules
  4. A model performs perfectly on training data
#11

What is the main idea of the K-Nearest Neighbors algorithm?

  1. It builds a sequence of decision trees
  2. It predicts using the closest training examples
  3. It calculates principal components
  4. It creates random clusters without distances
#12

How does KNN usually make a classification prediction?

  1. By averaging all features in the dataset
  2. By majority vote among the nearest neighbors
  3. By selecting the farthest neighbor
  4. By calculating a linear equation
#13

How does KNN usually make a regression prediction?

  1. By choosing the most frequent class
  2. By averaging the target values of nearest neighbors
  3. By creating a dendrogram
  4. By applying logistic regression
#14

What can happen when K=1 in KNN?

  1. The model can become sensitive to noise and overfit
  2. The model always becomes smoother
  3. The model ignores the training data
  4. The model becomes a Naive Bayes classifier
#15

What usually happens when the number of neighbors in KNN increases?

  1. The model becomes more complex and unstable
  2. The model becomes smoother and simpler
  3. The model stops using distance
  4. The model becomes a clustering algorithm
#16

Why is distance important in KNN?

  1. KNN does not use distances
  2. The nearest examples are selected based on distance
  3. Distance is used only in decision trees
  4. Distance replaces the target variable
#17

Why is feature scaling important for KNN?

  1. KNN works only with unscaled categorical data
  2. Features with larger scales can dominate distance calculations
  3. Scaling removes the need for training data
  4. Scaling converts regression into classification
#18

Which is an advantage of KNN?

  1. It never requires data preprocessing
  2. It is simple and easy to interpret
  3. It always works best with thousands of features
  4. It does not require a value of K
#19

Which is a limitation of KNN?

  1. It cannot use numerical data
  2. It may work poorly with very high-dimensional data
  3. It cannot be used for regression
  4. It does not depend on training examples
#20

What does a linear regression model use to make predictions?

  1. Only random cluster labels
  2. A weighted sum of features plus an intercept
  3. Only the nearest neighbor
  4. A dendrogram of classes
#21

In a linear model, what do weights represent?

  1. The influence of features on the prediction
  2. The number of rows in the dataset
  3. The number of clusters
  4. The size of the test set
#22

For one input feature, the prediction of a linear regression model is represented as:

  1. A circle
  2. A straight line
  3. A decision tree
  4. A cluster center only
#23

What does ordinary least squares aim to do in linear regression?

  1. Maximize classification error
  2. Find coefficients that minimize prediction error on training data
  3. Create clusters of equal size
  4. Remove all numerical features
#24

What is the main purpose of Ridge Regression?

  1. To increase the complexity of a model without control
  2. To reduce overfitting by shrinking coefficients
  3. To create cluster labels
  4. To remove the intercept from the model
#25

What happens when the alpha parameter in Ridge Regression increases?

  1. Coefficients become completely random
  2. Coefficients are pushed closer to zero
  3. The number of classes increases
  4. The model becomes K-Means
#26

How can Ridge Regression improve generalization?

  1. By using no training data
  2. By limiting model complexity through regularization
  3. By creating more target labels
  4. By fitting the test set separately
#27

What is a key property of Lasso Regression?

  1. It can make some coefficients exactly zero
  2. It never performs feature selection
  3. It is used only for clustering
  4. It always uses all features equally
#28

Why can Lasso be useful when there are many features?

  1. It increases all coefficients equally
  2. It can select only some important features
  3. It removes the target variable
  4. It converts numerical values into text
#29

Which statement is correct about Ridge and Lasso?

  1. Both always set all coefficients to zero
  2. Ridge shrinks coefficients, while Lasso can set some coefficients to exactly zero
  3. Ridge is only for clustering, while Lasso is only for PCA
  4. Neither method uses regularization
#30

Logistic Regression is mainly used for which type of task?

  1. Classification
  2. Clustering
  3. Dimensionality reduction
  4. Image reconstruction only
#31

In linear classification, how is a class prediction commonly made?

  1. By calculating only the average of all labels
  2. By applying a threshold to a linear function of features
  3. By using a dendrogram
  4. By randomly choosing the nearest cluster
#32

LinearSVC is an example of which type of model?

  1. Unsupervised clustering model
  2. Linear classification model
  3. Dimensionality reduction model
  4. Feature scaling model
#33

In Logistic Regression and LinearSVC, what does parameter C control?

  1. The number of clusters
  2. The strength of regularization
  3. The number of principal components
  4. The size of the dataset
#34

What does a small value of C usually mean in linear classification models?

  1. Stronger regularization and a simpler model
  2. Weaker regularization and a more complex model
  3. More clusters
  4. No training process
#35

What does a large value of C usually mean in linear classification models?

  1. Stronger regularization
  2. Weaker regularization and a more complex model
  3. No use of features
  4. Automatic dimensionality reduction
#36

What is multiclass classification?

  1. Predicting only yes or no labels
  2. Predicting one class from more than two possible classes
  3. Predicting a continuous number
  4. Grouping unlabeled data only
#37

Which task is an example of binary classification?

  1. Predicting house price
  2. Classifying an email as spam or not spam
  3. Grouping users into five clusters
  4. Reducing 30 features to 2 components
#38

Which is an advantage of linear models?

  1. They are fast in training and prediction
  2. They always outperform all other models
  3. They do not need input features
  4. They work only for image data
#39

Why are SGDClassifier and SGDRegressor useful?

  1. They are only for small datasets
  2. They are scalable versions of linear models
  3. They do not require optimization
  4. They replace all clustering algorithms
#40

What is the main idea of Naive Bayes classifiers?

  1. They consider each feature separately under a simplifying independence assumption
  2. They always build many decision trees
  3. They require no training data
  4. They only perform dimensionality reduction
#41

GaussianNB is most suitable for which type of data?

  1. Only binary data
  2. Continuous numerical data
  3. Only image pixels with no labels
  4. Only cluster labels
#42

BernoulliNB is commonly suitable for which type of data?

  1. Binary features
  2. Continuous regression targets
  3. Hierarchical clusters
  4. Principal components only
#43

MultinomialNB is often used for:

  1. Continuous sensor values only
  2. Count-based or discrete data such as text word counts
  3. Decision tree pruning
  4. K-Means initialization
#44

What is a decision tree?

  1. A model that makes predictions through a sequence of decision rules
  2. A model that only scales data
  3. A model that finds only principal components
  4. A model that cannot be visualized
#45

DecisionTreeClassifier is used for:

  1. Classification tasks
  2. Only clustering tasks
  3. Only PCA visualization
  4. Only feature scaling
#46

DecisionTreeRegressor is used for:

  1. Class label prediction only
  2. Predicting continuous target values
  3. Scaling features to 0 and 1
  4. Calculating Naive Bayes probabilities
#47

Why can decision trees overfit easily?

  1. They are always too simple
  2. They can create very detailed rules for training data
  3. They cannot split data
  4. They do not use target values
#48

What is pre-pruning in decision trees?

  1. Stopping tree growth early to reduce overfitting
  2. Removing the target variable
  3. Scaling data before training
  4. Increasing the number of trees
#49

What is post-pruning in decision trees?

  1. Training a tree without data
  2. Removing or reducing uninformative nodes after building the tree
  3. Converting a tree into PCA
  4. Choosing the nearest neighbor
#50

What does feature importance show in tree-based models?

  1. The number of missing values only
  2. How useful each feature is for prediction
  3. The number of clusters
  4. The number of test samples
#51

What is an ensemble method?

  1. A method that uses only one simple model
  2. A method that combines multiple models to improve performance
  3. A method used only for scaling
  4. A method that removes training data
#52

What is the main idea of Random Forest?

  1. To use one linear model
  2. To combine many decision trees
  3. To reduce data to two dimensions
  4. To calculate only Euclidean distance
#53

Why is Random Forest often more stable than a single decision tree?

  1. It ignores all weak features
  2. It averages or combines predictions from many trees
  3. It uses no training labels
  4. It performs only clustering
#54

What does the n_estimators parameter usually define in Random Forest?

  1. The number of features
  2. The number of trees
  3. The number of classes
  4. The number of clusters
#55

What is the purpose of n_jobs=-1 in Random Forest?

  1. To use only one CPU core
  2. To use all available CPU cores for parallel computation
  3. To remove all trees
  4. To disable training
#56

Which is a limitation of Random Forest?

  1. It cannot handle classification tasks
  2. It may work poorly on very high-dimensional sparse data such as text
  3. It always requires no memory
  4. It cannot estimate feature importance
#57

What is the main idea of Gradient Boosting?

  1. Build many independent trees without order
  2. Build trees sequentially so each new tree corrects previous errors
  3. Use only the closest neighbor
  4. Scale training and test data separately
#58

How does Gradient Boosting differ from Random Forest?

  1. Gradient Boosting does not use trees
  2. Gradient Boosting builds trees sequentially, while Random Forest builds many trees more independently
  3. Random Forest is always unsupervised
  4. Gradient Boosting is only used for PCA
#59

Why does Gradient Boosting require careful parameter tuning?

  1. Because model complexity and overfitting must be controlled
  2. Because it has no parameters
  3. Because it cannot learn from labels
  4. Because it only works with binary features
#60

What is the purpose of data preprocessing in machine learning?

  1. To prepare data for better model training and evaluation
  2. To delete all target labels
  3. To replace machine learning algorithms
  4. To create random answers
#61

What does feature scaling do?

  1. Removes the need for features
  2. Changes feature values to comparable ranges
  3. Creates new class labels
  4. Builds a decision tree automatically
#62

What does MinMaxScaler usually do?

  1. Converts all values into class labels
  2. Scales features to a fixed range, often from 0 to 1
  3. Removes outliers automatically
  4. Creates clusters
#63

What does StandardScaler usually do?

  1. Scales features using only the test set
  2. Transforms features to have zero mean and unit variance
  3. Creates decision boundaries
  4. Converts regression into classification
#64

Why should a scaler be fitted only on the training data?

  1. To avoid using information from the test data
  2. To make the test data easier to memorize
  3. To remove the need for a model
  4. To force all test values to be zero
#65

Why is it wrong to fit a scaler separately on the test data?

  1. It improves evaluation fairness
  2. It changes the test data using its own statistics and makes evaluation inconsistent
  3. It is required for all algorithms
  4. It increases the number of classes
#66

Why can scaling improve SVM performance?

  1. SVM ignores feature values
  2. Because SVM can be sensitive to feature scales
  3. Because scaling creates more labels
  4. Because SVM is a clustering algorithm
#67

What is dimensionality reduction?

  1. Increasing the number of features
  2. Reducing the number of features while preserving important information
  3. Removing all data points
  4. Changing classification into regression
#68

What is PCA mainly used for?

  1. Dimensionality reduction and visualization
  2. Supervised classification only
  3. Decision tree pruning
  4. Text tokenization only
#69

What information does PCA use when finding components?

  1. Class labels only
  2. Correlations and variance structure in the data
  3. Random cluster numbers
  4. Only the target variable
#70

Does PCA use class labels when finding principal components?

  1. Yes, always
  2. No, PCA is an unsupervised method
  3. Only for binary classification
  4. Only for logistic regression
#71

Principal components in PCA are:

  1. Original features without changes
  2. New directions formed as combinations of original features
  3. Class labels
  4. Decision tree nodes
#72

Why can PCA components be difficult to interpret?

  1. They contain no numerical values
  2. They are combinations of many original features
  3. They are always class labels
  4. They are created only from test data
#73

Why are the first two PCA components often used?

  1. To delete the dataset
  2. To visualize high-dimensional data in two-dimensional space
  3. To build a Random Forest
  4. To calculate K in K-Means
#74

In image analysis, PCA can be used to:

  1. Extract important patterns and reduce image dimensionality
  2. Create target labels manually
  3. Ignore pixel values
  4. Replace all classifiers permanently
#75

What is whitening commonly related to in PCA preprocessing?

  1. Making all images white
  2. Transforming components to have comparable scale and reduced correlation
  3. Removing the training set
  4. Changing class labels into numbers
#76

What is the main purpose of NMF?

  1. Supervised classification
  2. Feature extraction using non-negative components
  3. Creating decision trees
  4. Scaling values to negative ranges
#77

In NMF, coefficients are required to be:

  1. Negative
  2. Non-negative
  3. Only binary labels
  4. Random strings
#78

Why can NMF be useful for image or signal data?

  1. It can identify meaningful parts or components
  2. It requires class labels
  3. It always produces decision boundaries
  4. It removes all useful information
#79

What is the main purpose of t-SNE?

  1. Regression prediction
  2. Tree pruning
  3. Visualization of high-dimensional data in low-dimensional space
  4. Replacing feature scaling
#80

What information does t-SNE mainly use?

  1. Only target labels
  2. Distances or similarities between data points
  3. Only decision tree depth
  4. Only class names
#81

What is clustering?

  1. Predicting a known target variable
  2. Grouping similar data points without using labels
  3. Calculating regression coefficients
  4. Training a logistic regression model
#82

In clustering, what does a cluster label represent?

  1. The true supervised target
  2. The group assigned to a data point by the clustering algorithm
  3. The regression coefficient
  4. The model accuracy
#83

What does K-Means clustering do?

  1. Predicts continuous values from labels
  2. Assigns points to clusters based on nearest cluster centers
  3. Builds a Naive Bayes classifier
  4. Finds class probabilities
#84

What does K represent in K-Means?

  1. Number of features
  2. Number of clusters
  3. Number of test samples
  4. Number of principal components
#85

What are cluster centers in K-Means?

  1. Decision tree leaves
  2. Representative points of clusters
  3. Class probabilities
  4. Regression target values
#86

Which is a limitation of K-Means?

  1. It never requires choosing the number of clusters
  2. It can struggle with complex cluster shapes
  3. It cannot use numerical features
  4. It is always supervised
#87

Why can random initialization affect K-Means results?

  1. K-Means never uses initial centers
  2. Different initial centers can lead to different final clusters
  3. Initialization removes all features
  4. Initialization changes labels into probabilities
#88

In the context of K-Means, vector quantization means:

  1. Representing data points using cluster centers
  2. Training a logistic regression model
  3. Calculating class probabilities
  4. Removing all clusters
#89

What is agglomerative clustering?

  1. A bottom-up hierarchical clustering method
  2. A supervised classification algorithm
  3. A linear regression method
  4. A feature scaling technique
#90

What does hierarchical clustering produce?

  1. A logistic regression equation
  2. A tree-like structure showing relationships between clusters
  3. Only one regression coefficient
  4. A confusion matrix only
#91

What is a dendrogram used for?

  1. Visualizing the hierarchy of clusters
  2. Showing logistic regression coefficients
  3. Scaling numerical data
  4. Testing Random Forest parallelism
#92

What type of clustering algorithm is DBSCAN?

  1. Linear model based clustering
  2. Density-based clustering algorithm
  3. Supervised regression algorithm
  4. Tree ensemble classifier
#93

What can DBSCAN identify in addition to clusters?

  1. Regression coefficients
  2. Noise or outlier points
  3. Principal components
  4. Feature importance from trees
#94

In DBSCAN, what does epsilon define?

  1. Number of trees
  2. Neighborhood radius around a point
  3. Number of classes
  4. Learning rate of logistic regression
#95

In DBSCAN, what does minPts define?

  1. Maximum number of features
  2. Minimum number of points required to form a dense region
  3. Number of PCA components
  4. Number of decision tree levels
#96

Why is AI considered an infrastructure layer in modern information systems?

  1. Because AI only replaces hardware
  2. Because AI can be embedded into databases, workflows, pipelines, and decision processes
  3. Because AI removes the need for governance
  4. Because AI works only as a standalone chatbot
#97

What is a key characteristic of agentic information systems?

  1. They can only store files
  2. They can plan and perform multi-step tasks with some autonomy
  3. They do not interact with users or tools
  4. They are only used for static reports
#98

Why is human-in-the-loop control important in agentic AI systems?

  1. To remove accountability
  2. To supervise high-risk decisions and manage unexpected behavior
  3. To prevent all automation
  4. To delete audit logs
#99

What is the purpose of MLOps?

  1. To avoid deploying models
  2. To manage model training, deployment, monitoring, versioning, and validation
  3. To replace all databases with spreadsheets
  4. To remove data governance
#100

What is model drift?

  1. A model becoming physically damaged
  2. A change in data or environment that reduces model performance over time
  3. A method for clustering images
  4. A type of feature scaling
#101

What does Cloud 3.0 emphasize?

  1. Only local storage
  2. Integrated services, real-time responsiveness, hybrid environments, and platform engineering
  3. Removing APIs from systems
  4. Using no cloud services at all
#102

What is a hybrid cloud environment?

  1. A system using only personal laptops
  2. A combination of on-premises, edge, and cloud resources
  3. A model with no security controls
  4. A database without networking
#103

What is a data fabric intended to provide?

  1. Only manual data entry
  2. Consistent metadata, lineage, and access control across data sources
  3. Random model predictions
  4. Removal of all databases
#104

What is the purpose of FinOps in cloud environments?

  1. To remove cloud monitoring
  2. To manage and optimize cloud costs through observability and accountability
  3. To train decision trees
  4. To perform PCA on images
#105

What is the main goal of data governance?

  1. To make data unavailable
  2. To ensure data quality, security, compliance, lineage, and trustworthy use
  3. To delete metadata
  4. To avoid access control
#106

What does data lineage describe?

  1. The number of clusters in a dataset
  2. The origin, movement, and transformation history of data
  3. The depth of a decision tree
  4. The number of AI agents
#107

What is the benefit of real-time analytics?

  1. It delays all decisions
  2. It enables decisions based on current data at the moment they matter
  3. It removes the need for data streams
  4. It only works without automation
#108

What is an event-driven architecture useful for?

  1. Only static yearly reporting
  2. Reacting to changes and events as they occur
  3. Removing communication between systems
  4. Avoiding data integration
#109

What is a feature store used for in machine learning systems?

  1. To store only raw images
  2. To manage and serve model input features consistently
  3. To replace model evaluation
  4. To delete training datasets
#110

What is continuous monitoring in cybersecurity?

  1. Checking systems once a year
  2. Ongoing observation of activity to detect suspicious behavior
  3. Removing logs after deployment
  4. Disabling identity controls
#111

What is the main idea of Zero Trust security?

  1. Trust every internal user automatically
  2. Never trust implicitly and verify every access request
  3. Remove authentication
  4. Allow unlimited access to all systems
#112

What does the principle of least privilege mean?

  1. Give every user administrator rights
  2. Give users and services only the access they need
  3. Disable all permissions
  4. Share one password across all systems
#113

How can attackers misuse AI?

  1. AI cannot be used by attackers
  2. For automated phishing, imitation, and model attacks
  3. Only for improving system backup
  4. Only for scaling databases
#114

What is model poisoning?

  1. Improving model accuracy with clean data
  2. Manipulating training data or learning process to harm model behavior
  3. Compressing images with PCA
  4. Scaling features to zero mean
#115

What is a membership inference attack?

  1. A method for increasing cluster count
  2. An attack that tries to determine whether a specific record was used in model training
  3. A decision tree pruning method
  4. A type of linear regression
#116

What is red teaming in AI and cybersecurity?

  1. Deleting model documentation
  2. Testing systems by simulating adversarial attacks
  3. Training only unsupervised models
  4. Scaling data using MinMaxScaler
#117

What is confidential computing designed to protect?

  1. Only printed documents
  2. Data in use through secure hardware-backed environments
  3. Only public webpages
  4. Only cluster centers
#118

What is a domain-specific AI model?

  1. A model trained for no particular task
  2. A model optimized for a specific industry or type of problem
  3. A model used only for random numbers
  4. A model without evaluation
#119

What is an AI-native architecture?

  1. A system that avoids all models
  2. An architecture built around model runtimes, data pipelines, and continuous evaluation
  3. A paper-based workflow
  4. A single isolated spreadsheet
#120

What can personal AI agents do?

  1. Only turn off computers
  2. Perform multi-step personal or professional tasks such as scheduling and summarization
  3. Only create cluster labels
  4. Only run Ridge Regression
#121

What does Agent-as-a-Service mean?

  1. Selling only hardware servers
  2. Providing autonomous agent capabilities as a service
  3. Removing automation from enterprise systems
  4. Using only manual workflows
#122

Why is AI governance important?

  1. To prevent any model from being evaluated
  2. To ensure accountability, safety, compliance, and responsible model use
  3. To remove audit trails
  4. To hide model behavior from users
#123

What is explainability in AI systems?

  1. Making models impossible to understand
  2. Providing understandable reasons or insights about model decisions
  3. Removing all model outputs
  4. Increasing only training speed
#124

What is a model registry used for?

  1. Deleting old datasets
  2. Tracking model versions, metadata, and deployment status
  3. Replacing all databases
  4. Creating random forests automatically
#125

Why are audit trails important in AI-powered systems?

  1. They hide system behavior
  2. They record actions and decisions for accountability and investigation
  3. They remove user permissions
  4. They replace model training
#126

What is technical debt?

  1. A financial loan for buying computers
  2. Accumulated complexity or weak design choices that make future changes harder
  3. A feature scaling method
  4. A type of classification label
#127

What is the role of platform engineering?

  1. To prevent developers from using tools
  2. To provide reusable internal platforms and building blocks for teams
  3. To remove all cloud services
  4. To manually label every dataset
#128

Why can legacy systems be a challenge for innovation?

  1. They always use the newest architecture
  2. They can be difficult to integrate, scale, and modernize
  3. They never contain business data
  4. They automatically support all AI models
#129

Why is workflow redesign important for AI adoption?

  1. AI gives no value in workflows
  2. Major productivity gains often require redesigning processes around automation and agents
  3. It removes the need for monitoring
  4. It prevents human oversight
#130

What is the ideal role of humans in many AI-powered workflows?

  1. Humans should never supervise AI
  2. Humans supervise, audit, set policy, and handle exceptions
  3. Humans only provide random inputs
  4. Humans remove all governance rules
#131

How can agentic AI improve productivity?

  1. By slowing down all processes
  2. By automating multi-step routine workflows and reducing cycle time
  3. By removing all data
  4. By preventing tool access
#132

Why do AI agents need secure tool access?

  1. To bypass all permissions
  2. To perform actions safely while respecting identity, scope, and authorization limits
  3. To delete audit logs
  4. To avoid accountability
#133

Why is token lifecycle management important for agents?

  1. To make credentials permanent and unlimited
  2. To control how long and how broadly agents can access tools
  3. To remove identity checks
  4. To disable all APIs
#134

Why are rollback controls important in AI systems?

  1. To prevent updates forever
  2. To return to a previous safe state when a problem occurs
  3. To delete all model versions
  4. To avoid testing
#135

What is continuous evaluation of AI models?

  1. Testing a model only once before deployment
  2. Regularly checking model quality, safety, and performance over time
  3. Removing validation data
  4. Using only manual predictions
#136

What is bias in AI systems?

  1. A guaranteed improvement in accuracy
  2. Systematic unfairness or distorted behavior in model outputs
  3. A clustering center
  4. A scaling parameter
#137

What is privacy leakage in AI systems?

  1. Protecting all data perfectly
  2. Unintended exposure of sensitive information
  3. Increasing model explainability only
  4. Reducing dimensionality
#138

Why must sensitive data be carefully controlled in AI systems?

  1. Because it is never useful
  2. Because misuse can cause privacy, legal, and ethical risks
  3. Because it removes all model drift
  4. Because it prevents any prediction
#139

What does it mean to treat data as a product?

  1. Data is ignored after collection
  2. Data is managed with quality, ownership, documentation, and service expectations
  3. Data is never reused
  4. Data is stored without metadata
#140

What can a data SLA define?

  1. Only the color of charts
  2. Expectations for data quality, latency, and availability
  3. The number of hidden layers in a neural network
  4. The number of clusters in K-Means
#141

What is data provenance?

  1. The final accuracy score only
  2. Information about where data came from and how it was produced
  3. The number of CPU cores used
  4. A type of Naive Bayes classifier
#142

What are tamper-evident logs used for?

  1. To hide unauthorized changes
  2. To make unauthorized changes detectable
  3. To remove all monitoring
  4. To train PCA models
#143

What is a service mesh useful for?

  1. Only drawing charts
  2. Managing secure communication between services
  3. Removing APIs
  4. Replacing all databases with files
#144

What is edge computing?

  1. Processing data only in a distant central server
  2. Processing data closer to where it is generated
  3. Deleting local devices
  4. Using no network connection at all
#145

Why might latency-sensitive systems use edge or hybrid architecture?

  1. To increase delay
  2. To reduce response time by processing closer to the source
  3. To remove all security
  4. To avoid data collection
#146

Why is cloud cost observability important?

  1. To make spending invisible
  2. To monitor and control cloud expenses continuously
  3. To remove billing responsibility
  4. To disable elastic capacity
#147

What is a risk of poorly governed automation?

  1. It always improves accountability
  2. It can cause errors at scale without proper supervision
  3. It removes all operational complexity
  4. It prevents data leakage automatically
#148

What is important for successful enterprise AI adoption?

  1. Using models without monitoring
  2. Combining platforms, governance, security, data quality, and reskilling
  3. Avoiding all cloud services
  4. Ignoring model performance after deployment
#149

What does the idea of AI models as operating systems suggest?

  1. Models only write documents
  2. Models may orchestrate tools, applications, data access, and workflows
  3. Models cannot interact with services
  4. Models replace all hardware
#150

Which practice best supports responsible AI?

  1. Deploying models without testing
  2. Monitoring, explainability, governance, privacy protection, and human oversight
  3. Removing access controls
  4. Ignoring bias and drift