GARP Risk and AI (RAI) Practice Exam

These questions are part of the practice quiz. Start practicing

• Which discipline entails differentiating data types (quantitative vs qualitative) and applying controls for confidential and PII?
• Which method uses the entire training dataset to calculate the gradient for each update, ensuring a smooth path to the optimum but requiring significant memory?
• Adversarial testing is defined as?
• In regression modeling, what term denotes including higher-order terms to capture curvature in the relationship?
• Which approach replaces words with their dictionary form based on meaning and part of speech, turning 'better' into 'good'?
• Which term is more data driven and focused on making predictions?
• Agentic AI refers to?
• Which measure downweights common words to reflect rarity across a corpus?
• Which ensemble method builds trees sequentially, with each new tree focusing on the errors of previous trees?
• Which practice ensures all features are on the same scale to prevent domination by magnitude?
• Which learning paradigm is most suitable when you have some labeled data and a larger amount of unlabeled data?
• Which loss function punishes large errors more heavily than small errors?
• Which assumption states that unlabeled data naturally forms separable, locally dense clusters, implying nearby instances share the same label?
• If the Between-Clusters Sum of Squares relative to the total variance is high, what does this imply about cluster separation?
• What technique involves creating many slightly different versions of the data by randomly resampling?
• Which layer produces the final prediction or label?
• Basic RNNs struggle to remember information from far back due to which problem?
• Autoencoders are non-linear, encoder + decoder.
• Which approach is a semi-supervised method that predicts labels only for the unlabeled data at hand?
• Maintaining information about data origin, structure and definitions to support informed decision-making describes which data discipline?
• Which statement is true about common model metrics?
• In binary classification, which term describes a correctly identified negative case?
• Which term captures the risk of manipulating a model's outputs to influence user behavior?
• Which AI branch is focused on generating new content such as text, sound, or videos?
• If the base rates (prevalence of outcomes) differ between groups, it is mathematically impossible to satisfy demographic parity, predictive rate parity, and equal opportunity simultaneously.
• Which technique uses artificial neural networks to estimate values instead of storing them in a lookup table?
• In Weighted Least Squares, what is the purpose of the weights?
• Which description matches data collected over time for the same units, enabling analysis of changes over periods?
• The openness regarding a system's design, data, and algorithms.
• Which algorithm uses a specified K and a distance measure to classify a point by majority among its nearest neighbors?
• Which metric is the square root of Mean Squared Error, giving error in the original units?
• Which type of decision tree predicts a categorical outcome?
• A Regression Tree is a tree that predicts a continuous outcome.
• Which term describes adjusting variables to have a mean of zero and a variance of one, especially when outliers are present?
• Which classifier seeks to maximize the margin by identifying the optimal boundary?
• Which algorithm extends Q-learning by using a neural network to approximate the Q-values rather than storing them in a table?
• Which approach uses a simpler, interpretable model to approximate the behavior of a complex black-box model locally?
• Which metric is the square root of Mean Squared Error, expressed in the same units as the target variable?
• What is the name of the optimal boundary that is equidistant from the closest support vectors and maximizes the margin?
• Which prompting technique asks the model to think step by step to solve complex reasoning tasks?
• Which metric computes the average magnitude of errors, treating all errors equally?
• Which concept captures the possibility that systems may produce behavior not explicitly coded, leading to unforeseen outcomes?
• Which concept refers to focusing on a single metric rather than a broader view of model performance?
• Which measure assesses the expected cumulative future rewards from a given state, ignoring the specific actions taken?
• Ask many people the same question, then average their answers
• Morality based on adherence to ethical duties and rules, irrespective of consequences?
• The variance of the error term is not constant, which makes the model inefficient and masks the statistical importance of features.
• Which estimation method selects model parameters to maximize the probability of observing the given data under a presumed data-generating process?
• Which method is a value-based reinforcement learning method built on Temporal Difference learning, learning the value of taking an action in a state (Q-value) rather than just the state's value?
• Respecting the capacity of individuals to make their own informed, uncoerced decisions.
• Which risk category is most concerned with regulatory fines, lawsuits, or reputational damage arising from opaque AI decisions?
• Which classifier aims to draw the widest safety line between categories using only the most difficult examples?
• Partitioning the dataset by locating observations based on centroids (center points of clusters) is called what?
• Which technique adjusts the learning rate over time to prevent overshooting?
• Form of consequentialism that seeks to produce the 'greater net utility' for all affected?
• Which operation updates a cluster center by averaging all its assigned points?
• Which optimization method updates weights after seeing each individual, randomly selected data point?
• What are the adjustable parameters that scale the influence of input features on the neuron's activation?
• Which boosting variant is described as 'new trees learn from the remaining errors'?
• What component transforms the weighted sum to introduce non-linearity?
• Which criterion is explicitly designed to balance fit and complexity, with a lower score indicating a better model?
• Which learning approach combines labeled and unlabeled data, typically using the labeled data to predict pseudo-labels for the rest?
• Shapley Values quantify what in a model?
• Which graphical approach helps determine the optimal number of clusters by identifying an elbow in the plotted metric?
• Which search method divides the search space in half to locate an answer more quickly?
• Which term describes the measure used to determine similarity between word vectors?
• Which metric is defined as the area under the ROC curve (AUC)?
• Which strategy discovers new possibilities by exploring actions at random, often used in early learning?
• Which of the following describes the ROC Curve?
• Which representation records the number of times a word appears in a document?
• Which term is listed as a model adaption without a descriptive line in the material?
• Which clustering approach partitions a dataset by locating observations based on centroids?
• Which search strategy analyzes moves by considering the opponent's best counterplay (minimax)?
• What term denotes the constant added to the weighted sum?
• Which phrase describes a solution that seems good but isn't truly optimal, requiring overcoming a short dip?
• Which term describes reliance on AI outputs that feel plausible but may be wrong, leading to miscalibration of trust?
• In high-dimensional environments, which approach uses neural networks to estimate values instead of a tabular table?
• Which metric is defined as the proportion of real positive events that are captured?
• What is the smallest, most useful summary of this data?
• Which term denotes the family of models capable of performing tasks like translation and question answering, often trained on large datasets?
• Which clustering method creates a multi-level hierarchy of clusters?
• Which concept describes the situation where predictor variables are highly correlated, making it difficult to distinguish their effects?
• Which problem occurs when the learning signal becomes so small that earlier parts stop learning?
• Choosing which fairness measure to prioritize is a moral and ethical judgement rather than a purely technical decision.
• Which regularization uses an L1 penalty to drop inputs that are not useful?
• What is the iterative art of designing inputs to get the best possible results from an LLM?
• Which method constructs an ensemble by training on bootstrapped samples and aggregating predictions?
• Which clustering approach defines clusters as regions where density points exceed a threshold?
• Not organized
• Which metric expresses error as a percentage of actual values?
• Which metric is used to quantify how close two word vectors are in meaning?
• Which metric measures how wrong predictions are overall?
• Which term corresponds to the inclusion of non-linear terms to capture curvature or turning points in the relationship?
• In reinforcement learning, which strategy involves the agent selecting actions at random to explore new possibilities?
• Which metric is most fundamentally concerned with the proportion of actual positives captured by the model?
• Which term denotes incorrectly flagging a negative case as positive?
• Which boosting method weights misclassified data points more to focus learning on them?
• Individual Fairness means?
• Which factor is most directly affected by outliers in K-means centroid calculations?
• Demographic Parity criterion describes?
• Which data subset is used to determine the chosen model's effectiveness?
• What term describes the decay of essential human skills due to prolonged reliance on AI?
• What is Tokenization?
• Which of the following is a common challenge of neural networks?
• Which metric is used to quantify how mixed a group is, often used to measure impurity in decision trees?
• What data organization is described as data stored in rows and columns?
• Which quality metric assesses how close a data point is to its own cluster relative to its distance to the nearest other cluster?
• What model is loosely modeled on how the brain performs computation?
• Which term describes data that is not organized in any schema or format?
• Which type of decision tree predicts a continuous outcome?
• Which method systematically tests multiple combinations of hyperparameters to find the best model?
• Which type of hierarchical clustering starts with individual data points and groups them by similarity?
• Logarithmic transformations are particularly helpful when which condition holds?
• Which measure evaluates how good a specific state is, in terms of expected future rewards?
• Which phenomenon involves using models to target advertising or influence politically?
• Which term denotes a correctly identified negative instance?
• Which method updates estimates step-by-step while the episode is in progress, using bootstrap estimates?
• Which pruning method uses a separate validation set to prune branches?
• Which approach uses a parameter to determine whether to explore or exploit at a given time?
• Which hierarchical clustering method begins with all data points in one cluster and splits them into smaller clusters?
• Which method changes the shape of the data (not just its scale) to make it easier to analyze or model, especially when data are highly skewed?
• In reinforcement learning, what is the term for the entity that makes decisions?
• Which evaluation tool is described as the foundation for classifying outcomes by crossing predicted and actual labels?
• Which term describes data lying on a lower-dimensional manifold embedded in high-dimensional space?
• Which approach in word embedding training predicts surrounding context words from a center word?
• Which term designates the set of data points that lie closest to the boundary in a classifier?
• Which representation yields a numeric vector by counting word frequency across a document?
• Which method draws the best straight line to explain how one variable changes when another single variable changes?
• Which clustering limitation means that K-means cannot reliably identify clusters that are not spherical in shape?
• Which approach fills in the target word by using surrounding context words?
• Which model selection criterion is designed to assess the trade-off between goodness-of-fit and model complexity, with lower values preferable?
• To avoid overshooting as the model nears the optimum, the learning rate can be decayed over time using exponential or inverse functions.
• What term refers to the variables used to explain the target in a predictive model?
• What step converts the processed text into numerical vectors for model analysis?
• The fairness concept that ensures equal true positive rates across groups is called...
• Which technique effectively removes inputs by shrinking some coefficients to zero?
• What term describes the process of repeating allocation and recalculation until the centroids stop changing?
• Model users failing to recognize what a model cannot capture is best described as which issue?
• Which prompting method involves explicitly guiding reasoning steps before concluding?
• Which method uses a sample of documents labeled by humans to learn which words are associated with specific categories?
• Which approach ranks input features by their influence on the final result, without considering interactions?
• Which line provides independent assurance that the governance framework is operating effectively?
• Which concept refers to the openness regarding a system's design, data, and algorithms?
• A classic transductive semi‑supervised method.
• What evaluation method involves training and testing multiple times across different data partitions to assess generalization?
• Which function includes design and implementation of models, monitoring daily performance and investigating threshold breaches?
• Which search strategy involves assuming the opponent minimizes your chances and uses the minimax principle?
• Which metric measures overall accuracy across all predictions?
• Which term describes computing the new cluster center by averaging member points?
• In a decision tree, the very first question or top node is known as the
• Which approach focuses on finding the optimal policy that maps a given state to an action?
• Which prompting style provides examples?
• Goal of Econometrics which is to prove a theory.
• What is recommended to assess model performance beyond a single metric to ensure robust conclusions?
• Which criterion requires that precision of positive predictions is identical across groups?
• Which ethical framework emphasizes virtuous character and living a good life rather than following rules or maximizing outcomes?
• Which issue arises when the system cannot explain how it arrives at a decision, creating accountability gaps?
• What is the name of the learning method in neural networks that adjusts weights by propagating errors backward through the network?
• Which concept describes the issue where using flawed proxies (like healthcare spend as a proxy for sickness) can bias results toward wealthier groups?
• Unintentionally interpreting results to align with preconceived beliefs is known as which bias?
• Transductive methods focus only on the specific unlabeled data you already have.
• Which principle is central to data protection compliance?
• Data Transformation takes many correlated variables and replaces them with a smaller number of new variables that capture most of the information.
• Which term refers to systems that often fail in unexpected ways when applied beyond their narrow boundaries of intended use?
• Which term describes a generation method that limits token choices to a fixed number of top options at each step?
• Which clustering method starts with each data point as its own cluster and merges similar clusters?
• What is the F1 Score?
• Which approach is typically used to determine actions directly by optimizing a policy over actions in a given state?
• What are the training data points that lie closest to the classification boundary called?
• Occurs when the training data do not reflect the target population.
• Which method gives less weight to noisy observations and more weight to precise ones?
• Which term captures the power imbalance created by opaque AI models that limit external understanding?
• What term describes plausible but factually incorrect LLM outputs?
• Which regularization keeps all features but prevents any single feature from becoming too influential?
• Which distance metric is defined as the straight-line distance in Euclidean space between two points?
• Collecting only the data strictly necessary for a specific purpose.
• Which measure assesses a point's influence on parameter estimates in a regression model?
• Which term corresponds to replacing the surrounding system as a type of model adaption?
• Which term best describes the entire model composed of layers, weights, and activations used to model complex patterns?
• Which risk describes exposure to biased or unfair decision making and threats to personal privacy?
• What is the primary purpose of the Akaike Information Criterion (AIC) in model selection?
• Which scheme records the number of times that a word appears in a document?
• What are the layers between the input and output called, which process information within a neural network?
• Dimensionality Reduction identifies new directions in the data that explain the most variation. These directions are called?
• Which term denotes the terminal node that represents the final prediction?
• Which measure indicates how often a word appears in a specific document relative to that document's length?
• Which concept refers to a model that does not remember prior prompts, making each interaction independent?
• The "smart compression machine" refers to which model?
• In a quadratic model, the relationship between the feature and the target is no longer constant and differentiation is required to find the point where the target variable is maximized?
• In Principal Components Analysis, each new component is constructed to be uncorrelated with the previous ones.
• Normalization changes the data in addition to scale by altering the shape of the data to facilitate analysis, especially when skewed.
• Plotting WCSS against the number of clusters is used to identify the elbow; what is the method commonly referred to as?
• What is the focus of value-based reinforcement learning?
• Which area focuses on obtaining consent, data minimization, and protecting data subject rights?
• What is the maximum number of tokens the LLM can process in a single sequence?
• Don't rely on one opinion combine many opinions. So instead of trusting one tree, we: build many trees and combine their answers
• What technique replaces missing values with the mean or median of existing data?
• Which metric measures how surprised a language model is by a piece of text?
• Linear regression is easily distorted by outliers, missing data, and measurement errors. Which term describes this data sensitivity?
• What term describes the phenomenon where distance-based methods like K-means lose effectiveness as the number of features increases?
• Which problem occurs when the learning signal becomes too large, causing unstable updates?
• Which method chooses model parameters so predictions are as close as possible to actual data on average?
• Which method is the same idea as OLS but for nonlinear models?
• Which concept explains why distance-based clustering methods become less effective in high-dimensional spaces?
• Which penalty combines L1 and L2 penalties to shrink coefficients and drop useless inputs?
• The ability to describe how a system reached a decision.
• AI systems should provide transparent and understandable explanations for their decision.
• Which metric is defined by the average of squared errors, punishing larger errors more than smaller ones?
• Which type of hierarchical clustering starts with grouped data points and splits them by hierarchy?
• Which metric expresses error as a percentage of actual values?
• Which learning paradigm involves learning from labeled data to categorize data or predict numerical values?
• Which method keeps all features but shrinks coefficients toward zero and penalizes large coefficients to stabilize estimates?
• Which practice involves deleting data as soon as it is no longer needed and using encryption and anonymization to keep data safe?
• What term describes the point at which a variable no longer has an effect on results?
• Which practice emphasizes deleting data when unnecessary and using encryption or anonymization to keep data safe?
• What is the step-by-step method to update weights to minimize error?
• Which concept describes a mismatch between training data and the target population that can bias results?
• What term describes a long stretch where nothing seems to change despite continued effort, indicating stagnation in learning?
• What concept describes that linear regression assumes straight-line relationships, while real-life relationships can be curved or change direction?
• BLEU and ROUGE are what type of metrics?
• Which density-based clustering method defines groups as clusters and treats outliers as noise?
• Which technique traces errors backwards to adjust parameters?
• Linear regression is suitable for predicting continuous values, not categories. Which term describes this limitation?
• Which technique maps words to their base dictionary form using linguistic rules, often considering the word's part of speech?
• Which concept describes threats where complex information hidden in data that people wish to keep private?
• Let the tree grow fully, then remove weak parts
• Which method explains a decision by showing what would need to change for a different outcome?
• Which ensemble approach is built by repeatedly sampling data and training on those samples to improve stability?
• What term is added to the weighted sum to adapt the model to data?
• What concept describes the idea that poor data quality leads to poor model outputs?
• In a decision tree, the path you take after answering the question is called a
• Which statistical measure is used to compare and select models, especially in regression, where a lower number indicates a better model?
• Which term is associated with the residual defined as Actual minus Predicted?
• Which term best describes AI systems that perform actions autonomously?
• Which technique reduces variance by averaging predictions from multiple models trained on bootstrap samples?
• Which term refers to extreme values that distort results?
• When choosing to prioritize one fairness measure over another is more about moral judgment than technical computation, this trade-off is known as...
• Which term specifically evaluates how good a particular action is in a given state?
• Which policy category includes implementing permission, encryption, and data retention policies to prevent corruption or loss?
• Making it easy for individuals to withdraw consent, access their data, or delete it.
• Which principle is primarily concerned with preventing harm to others (non-harm)?
• Don't let a tree grow too large in the first place
• Which search method uses a rule of thumb to decide which option to explore first?
• Extreme data points exert a disproportionate influence on the regression line, often distorting results.
• Which approach uses a predefined list of words to calculate a net sentiment score?
• Which risk describes countries not able to leverage AI as quickly fall behind those who can?
• What is the primary role of the Model Inventory function?
• What coefficients are used for categorical data, representing average differences relative to a baseline?
• Developing a vision, setting goals, and establishing a Data Governance Committee for oversight corresponds to which data-management function?
• Group Fairness refers to?
• What is the straight-line distance between two points in space commonly called?
• Which term refers to randomly selecting the initial centroids in centroid-based clustering?
• In clustering, inertia is also called the measure that sums squared distances from points to their centroid. Which label correctly names this metric?
• Can you rebuild the original data from this summary?
• Which sampling method restricts the model to sampling from only the K most probable tokens at each step?
• Which concept describes a user not realizing something is wrong when the system provides a plausible answer with confidence?
• Also known as segmentation, separates data into groups where items in the same group are similar and items in different groups are dissimilar.
• Reinforcement learning aims to maximize which of the following?
• In centroid-based clustering, which step assigns each point to the nearest centroid based on a distance measure?
• Which principle requires data to be used in a way that respects the original context of its collection?
• Which risk category specifically addresses reputational and regulatory exposure when AI practices cannot be explained?
• Which technique squashes data into a fixed range between zero and one?
• What term denotes incorrectly labeling a positive case as negative?
• Which metric is minimized as a measure of how far predicted values are from observed values in regression?
• Which term refers to removing HTML and web-scraped data and expanding contractions?
• Which regularization method uses an L2 penalty to shrink coefficients but not drop inputs?
• What is the machine learning technique that develops a policy to maximize a long-term or cumulative reward through trial and error?
• How do RNNs conceptually process input?
• A way of studying relationships between economic variables by starting with a theory, choosing a mathematical model that represents that theory, and then checking if the real-world data supports it.
• Which term refers to the data points most critical to defining the decision boundary in SVMs?
• Which encoding technique creates binary indicators for each category in nominal data?
• Which term describes a generation method that uses the smallest set of tokens whose cumulative probability reaches a threshold?
• Which term describes the ability to describe which factors and logic led to a model's decision?
• Which term describes data issues where extreme values distort results?
• Which term denotes a correctly identified positive instance?
• Predictive Rate Parity is about?
• What term describes something that looks flat locally, even if it's curved globally?
• Which data subset is used for model selection and tuning hyperparameters?
• Which concept describes the tendency to trust automated systems blindly?
• Which term describes the difficulty of explaining AI decisions to stakeholders due to opacity?
• Which representation treats every word in a document as an independent feature, ignoring word order?
• Which issue arises when outliers significantly distort the position of centroids or create isolated clusters?
• Using personal data only in accordance with the norms of the context in which it was originally given.
• Which term describes a model that fits training data very closely but generalizes poorly to new data?
• Which concept refers to the situation where the search space becomes too large to search exhaustively?
• Which network is designed for sequences where order matters?
• Autoencoders learn by making mistakes and fixing them. The learning signal is the
• Which algorithm estimates the quality of action-state pairs and typically follows an off-policy learning framework?
• Which fairness criterion requires identical likelihood of a positive prediction across subpopulations regardless of correctness?
• What metric checks whether model output is identical to a reference answer character-for-character?
• Which technique measures the contribution of each input feature to the model's prediction across all possible feature combinations?
• Which encoding method is used for nominal data with no natural order, often creating binary indicators for each category?
• There is a smooth and continuous boundary separating classes. If two points are 'close' in the feature space, their labels should be the same.
• When comparing two models using a given criterion, a lower value indicates what?
• At each step, the agent has possible options called?
• Feature Importance Scores rank inputs by what's?
• Which estimation technique is used when the relationship between variables and parameters is nonlinear and you fit by minimizing the squared error in the predicted observations?
• Which phenomenon involves not relying on one's judgment because AI appears to be an expert?
• In the context of estimating a model, which concept is described as being used to estimate the model's weights and biases?
• Which category includes data types such as text, image, audio, video, and multimodal data?
• Which problem solving technique is based on breaking a complex problem into simpler subproblems?
• Which line provides final approval for frameworks and ensures alignment with strategic objectives?
• Which approach chooses parameters that maximize the probability of observing the data given a model?
• Which cognitive bias describes overreliance on AI leading to insufficient scrutiny?
• Which term describes theoretical concerns about superintelligent AI making decisions against human interests?
• Which term describes the difficulty of determining which parts of a model remain valid when the environment changes?
• Which penalty combines L1 and L2 penalties to shrink coefficients and drop useless inputs?
• Perfect group fairness often requires compromising overall accuracy.
• Which optimization method uses the entire training dataset to compute the gradient for each update?
• Which principle focuses on fairness, equality, and impartial distribution of benefits and burdens?
• Organized in rows and columns
• Which concept improves accuracy by anchoring the model to verified data sources through retrieval?
• Which activation function passes positives and blocks negatives, aiding learning inside the network?
• A risk described as model users not understanding underlying assumptions is best captured by which term?
• Which learning paradigm identifies patterns in unlabeled data, such as clustering?
• The agent's strategy or plan of action that maps states to actions is called a
• For datasets with labeled and unlabeled observations, primarily used for classification rather than prediction, which learning paradigm is appropriate?
• Which concept states that the future state depends only on the current state and action, not the past?
• Which example is used to illustrate image generation from a text prompt among content modalities?
• Which metric measures, among instances flagged positive, how many are actually positive?
• What is the term for the distance between the decision boundary and the closest data points from each class?
• Which data type has a natural order but the intervals between values are not necessarily equal?
• Which technique reduces many correlated variables to a smaller number of uncorrelated components that capture most of the information?
• Which data type has no inherent order and represents categories without rank?
• Morality judged solely by outcomes or results?
• Which moral obligation requires one to avoid harming others?
• Which statement best describes the purpose of word embeddings?
• Which component encodes the agent's plan of action, mapping states to actions?
• Even with good data collection, this can cause algorithms to underperform on statistical minorities.
• Which risk describes spread of deepfakes and misinformation affecting public opinion or elections?
• Which concept refers to the ability to understand why a system made a particular decision?
• What term describes the strategy where the agent always selects actions with the best observed rewards so far?
• Maintaining information about data origin, structure and definitions to support informed business decisions describes which data discipline?
• Which ensemble method builds multiple trees by training on bootstrapped samples to ensure diverse decisions?
• In multiple regression, what term describes the partial effect of a particular variable while holding other variables constant?
• What term refers to the connections between neurons that carry numbers and determine influence?
• Which phenomenon refers to erosion of human skills due to AI reliance?
• Retrieval Augmented Generation (RAG) enhances AI by combining generation with retrieval from external data sources to improve what?
• Which statement correctly describes the F1 Score?
• Reduces tree size to avoid overfitting and improve interpretability
• Which term describes the set of modeling techniques that include LLMs, Transformers, VAEs, and GANs for text and image generation?
• Which approach fills in missing values using the mean or median of available data?
• Which metric summarizes the ROC curve into a single number that indicates separation quality?
• Which variant records 1 if a word appears in a document and 0 otherwise?
• Which approach uses a center word to predict surrounding context words?
• The degree to which a human can comprehend the inherent logic and internal architecture of a model.
• Which technique reduces dimensionality by projecting data onto principal components?
• In PCA, what are the principal components?
• If a natural order exists among categories, which encoding approach should be used?
• In some clustering analyses, the analyst must pre-determine the number of clusters. Which description best matches this requirement?
• What term describes the difficulty outsiders have understanding the reasoning behind outputs produced by neural networks because of their opaque nature?
• What term denotes the variable we are trying to explain in a predictive model?
• What term describes methods for comparing relationships between words using vector operations?
• Can be continuous (income, age) or discrete (number of emails received).
• Which term refers to the presence of complex information hidden in data that people want to keep private?
• Which line of defense is described as setting standards, conducting independent validations, and maintaining the model inventory?
• What describes dense vector representations of words that capture meaning and relationships?
• Which criterion computes the sum of squared residuals between observed and predicted values?
• Which network is designed to address vanishing gradient by maintaining memory across time steps?
• Which term describes data collected over several time periods for the same units?
• Which risk describes potential job losses in specific fields or widened wealth gap?
• What is the umbrella term for methods that train multiple models and combine their outputs to improve predictive performance?
• Which metric is defined as the sum of squared distances between cluster centroids and the global centroid, indicating cluster separation?
• Which technique is used to learn nonlinear dimensionality reduction by training networks to reconstruct inputs?
• Verifying the legal right to use data is best described by which data management concept?
• Distinguishing between quantitative and qualitative data and applying controls for confidential and PII describes which concept?
• Which risk category concerns global disparities in AI adoption rates and capabilities?
• Which approach learns a policy directly without necessarily modeling a value function?
• Which regularization method tends to shrink coefficients toward zero without necessarily setting them to zero?
• Which term describes a model with high bias but low variance?
• Which parameter is used to gradually reduce epsilon over time, causing exploration to be heavy early and exploitation later?
• Which method extends simple linear regression by considering several causes at once to explain the outcome?
• Which assertion states that neighboring points in feature space are likely to share the same label?
• Which measure assesses how good a specific action is given a state?
• Which problem describes the difficulty of keeping track of which aspects of a situation stay the same or change when an action is performed?
• Which term describes the risk of over-reliance on AI leading to loss of human skills (as per the provided term)?
• What is the name of the approach that learns from labeled and unlabeled data to generalize to unseen data?
• What term describes the value the target would take if the feature were zero, representing bias in the model?
• Which moral obligation is to take positive steps to help others?
• Implied Natural Order
• Which evaluation approach involves testing the model multiple times on different data slices to obtain a robust performance estimate?
• Which term refers to data that groups information into distinct categories or labels rather than numerical values?
• Which step in centroid-based clustering assigns each data point to the nearest centroid?
• Which term is described as the process of grouping data into clusters where items in the same group are similar and items in different groups are dissimilar?
• Which concept describes a combinatorial explosion where the number of possible outcomes grows exponentially?
• Which assumption posits that high-dimensional data lies on lower-dimensional substructures called topological manifolds?
• Which term describes extreme observations that can bias regression results?
• Which activation function outputs values between zero and one and is common in classification?
• Which term describes a model with low bias but high variance?
• What pictorial representation shows the steps of hierarchical clustering, illustrating how clusters are merged?
• A Classification Tree is a tree that predicts a categorical outcome.
• Which problem arises when users focus on a single number rather than a range of potential outcomes or confidence intervals?
• Linear Discriminant Analysis for Classification classifies observations by
• Which classifier is based on assuming independence among features and combines probabilities?
• Standardization is preferred when outliers are present because it adjusts variables to what properties?
• Which method also penalizes coefficients, but can shrink some coefficients all the way to zero, performing feature selection?
• Which issue arises when analysts neglect the limitations of a model?
• Which concept refers to ensuring accuracy, consistency and integrity of data?
• Which term describes AI systems that primarily predict the next word rather than possessing true understanding?
• Which learning method updates estimates after an entire episode has finished?
• Which term describes a diagnostic tool used to analyze residuals and assess model adequacy by visual inspection?
• Which issue arises when features are highly correlated, making it difficult to disentangle their individual effects and causing unstable parameter estimates?
• Which term describes the architecture that enables attention across the entire input and parallel processing?
• Which method estimates the probability that something belongs to a particular class, and then uses a cutoff to make a yes/no decision?
• Which neural network is designed to work with images and spatial data?
• Which line is described as maintaining the model inventory and setting standards for validation?
• Adversarial testing (red-teaming) is best described as?
• Which type of model adaption involves updating code or interfaces?
• Inductive methods try to learn a general rule that can be reused on new data.
• Which concept describes bias in training data that can be learned and perpetuated by models?
• Which metric compares an observation's distance to its own cluster (cohesion) with its distance to the nearest neighboring cluster (separation)?
• In reinforcement learning, what is the term for the signal indicating the desirability of an action's outcome in the present step?
• Another way to measure how mixed a group is
• Which learning approach uses trial-and-error to develop policies for a sequence of decisions?
• Which term describes the idea of using non-linear forms to model relationships beyond straight lines?
• Which fairness concept focuses on ensuring that individuals who are truly positive cases have an equal chance of being correctly identified across demographic groups?
• Which bias causes results to be interpreted to support preconceived beliefs?
• Which concept states that missingness itself is related to the outcome?
• Which model is designed to remember important information for a long time and forget unimportant information?
• In a decision tree, a place where a question is asked is called a
• The right to have private information removed from directories when it is inadequate or irrelevant.
• Which density-based method defines similarity by overlap in nearest neighbors rather than raw distance?
• Which component applies a non-linear transformation to a neuron's weighted input?
• Which concept describes over-reliance on AI that makes people accept AI judgments as correct without critical appraisal?
• BLEU and ROUGE are categorized as which type of metrics?
• The immediate feedback after an action is called what?
• Data at a specific moment in time is described as...
• Which issue occurs when the modeling process fails to account for outcomes the model cannot represent?
• Which concept describes unanticipated AI system responses arising from complex interactions?
• Which strategy combines exploration and exploitation by choosing a random action with probability epsilon and the best known action otherwise?
• Which term describes a predictive, data-driven approach used for forecasting?
• Which process reduces words to their base or root form by removing prefixes and suffixes?
• Which property describes LLMs that do not retain information from one prompt to the next?
• Which method updates weights after seeing each individual, randomly selected data point?
• Which term describes model users not understanding underlying assumptions?
• What is the process of collecting, cleaning, and visualizing data before building a model?
• Which concept describes performance degradation due to changing environmental factors?
• In a decision tree, the final predicted value or category is located at the
• Which preprocessing step removes common words with no informational value like 'a' and 'the'?
• What do the model's coefficients measure in a multiple regression context?
• Which term names the traditional approach to AI based on explicit symbolic manipulation and rule-based reasoning?
• Which term describes training data biases that can be learned and perpetuated by models?
• A form of post-pruning
• Which risk category is primarily concerned with misinformation campaigns and deceptive content affecting public opinion?
• Which outcome is described by reduced vigilance in critical scenarios leading to safety failures?
• Which sampling method selects the smallest set of tokens whose cumulative probability reaches the threshold?
• Which data type has no natural order?
• Which term is used for the initial stage of data work focused on understanding data before modeling?
• Which term describes the architecture that processes the entire input sentence at once, enabling attention across the full sequence?
• Each principal component is a straight-line (linear) combination of the original features?
• Cost complexity pruning
• The duty to act in accordance with fairness, equality, and impartiality.
• Which dimensionality reduction technique is linear, mathematical, and assumes data lie on a linear subspace?
• What foundational tool summarizes the performance of a classifier by comparing predicted vs actual outcomes?
• What is the term for the predicted change in the target variable for a one-unit change in the feature?
• Which term best describes a system's susceptibility to adversarial examples that humans would recognize as incorrect?
• Which term describes data described by many measurements at once?
• Which prompting style gives no examples?
• Which cross-validation technique ensures each fold preserves the overall distribution of the target variable?
• Instead of all data (batch) or one item (stochastic), you use small chunks.
• Which impurity measure is commonly used in CART and does not rely on logarithms?
• Which right allows private information to be removed if inadequate or irrelevant?
• Which parameter changes the shape of the probability distribution over next tokens?
• Which risk describes exposure to reputational, regulatory and legal risk when AI practices hurt a group or can't be explained?
• Which term is characterized by explicit, rule-governed processing in artificial intelligence?
• What do you call the parameters you set in the model before training rather than the ones learned by the algorithm?
• Which concept explains how outliers, missing data, or measurement errors can distort linear regression results?
• What term describes the environment or circumstances in which decisions are made?
• In reinforcement learning, which problem involves choosing among multiple arms with uncertain rewards to maximize cumulative gain?
• Which term refers to the absence of accountability due to opaque decision-making in AI systems?
• Which concept captures the idea of evaluating moves under the assumption of optimal counter-moves by the opponent?
• Which technique groups words into a common base form based on meaning and part of speech?
• Which function maintains a central registry of model usage, approvals, and limitations?
• Which concept describes models with billions of weights that are difficult for humans to interpret, creating a black-box system?
• What is the topmost node of a decision tree called?
• Which limitation describes LLMs making probabilistic guesses rather than understanding absolute truth?