D429 Introduction to AI for Computer Scientists - Set 1 - Part 1

Correct Answer: Learning agent system

Option b is correct because learning agents can adapt to new medical patterns, improve decision-making through feedback, and handle the complexity of medical prioritization effectively.
Option a is incorrect because simple reflex agents only respond to current percepts without considering history or learning from past decisions in critical healthcare scenarios.
Option c is incorrect because model-based reflex agents maintain state but cannot improve their performance over time through experience with patient outcomes.
Option d is incorrect because utility-based agents optimize for utility but lack the adaptive learning capabilities essential for evolving medical knowledge.

Correct Answer: Event calculus fluents

Option b is correct because event calculus with fluents can represent both temporal changes in assembly procedures and dynamic relationships between components during manufacturing.
Option a is incorrect because semantic networks alone cannot adequately represent the temporal aspects and changing states required for sequential assembly line procedures.
Option c is incorrect because modal logic focuses on necessity and possibility rather than the temporal sequences and state changes in assembly processes.
Option d is incorrect because description logics emphasize static categorical definitions rather than dynamic temporal relationships needed for assembly line variations.

Correct Answer: LSTM network architecture

Option c is correct because LSTM networks have memory cells and gates specifically designed to maintain long-range dependencies and context across extended conversations.
Option a is incorrect because bag-of-words models treat text as unordered collections, completely losing the sequential context essential for understanding conversation flow.
Option b is incorrect because one-hot encoding only represents categorical data as binary vectors without any mechanism for maintaining conversational context.
Option d is incorrect because tokenization is just a preprocessing step that converts text to tokens without any context preservation capabilities.

Correct Answer: Partially observable stochastic

Option b is correct because fog creates partial observability where sensors miss information, and traffic behavior remains unpredictable, requiring probabilistic decision-making approaches.
Option a is incorrect because heavy fog prevents full observation of the environment, and vehicle movements cannot be deterministically predicted in traffic.
Option c is incorrect because the vehicle still has some sensor data available despite fog, making it partially rather than completely unobservable.
Option d is incorrect because semi-dynamic refers to environmental changes only when agents act, not the observability and uncertainty issues from fog.

Correct Answer: Weighted loss functions

Option b is correct because weighted loss functions specifically address unbalanced classes by giving more importance to minority class errors during model training.
Option a is incorrect because clustering algorithms group similar data without specifically addressing the class imbalance problem in fraud detection scenarios.
Option c is incorrect because unsupervised learning lacks the labeled fraud examples needed to distinguish between fraudulent and legitimate transaction patterns.
Option d is incorrect because passive reinforcement learning observes policies without actively addressing the fundamental class imbalance issue in the dataset.

Correct Answer: Inverse reinforcement learning

Option b is correct because inverse reinforcement learning specifically learns reward functions and preferences by observing behavior without explicit reward signals.
Option a is incorrect because supervised classification requires labeled training data with explicit categories rather than inferring preferences from observed behavior.
Option c is incorrect because active reinforcement learning requires explicit rewards and actively explores actions rather than learning from observations.
Option d is incorrect because regression analysis predicts continuous values but cannot infer underlying preference structures from behavioral observations alone.

Correct Answer: Bayesian inference networks

Option b is correct because Bayesian networks provide probabilistic reasoning with clear causal relationships that doctors can interpret and verify against medical knowledge.
Option a is incorrect because deep neural networks often function as black boxes, making it difficult to explain diagnostic reasoning to healthcare professionals.
Option c is incorrect because black box models by definition lack interpretability, which is essential for medical professionals to trust and verify diagnoses.
Option d is incorrect because unstructured data processing is a data handling approach, not a technique for providing interpretable diagnostic reasoning.

Correct Answer: Federated learning approach

Option b is correct because federated learning trains models across distributed devices while keeping data local, preserving privacy and meeting regulatory requirements.
Option a is incorrect because data augmentation creates variations of existing data but doesn't address the privacy concerns of centralizing sensitive customer information.
Option c is incorrect because feature extraction alone doesn't solve the problem of training models without centralizing the sensitive raw customer data.
Option d is incorrect because centralized database storage directly contradicts the requirement to keep customer data distributed and locally secured.

Correct Answer: Anomaly detection clustering

Option b is correct because anomaly detection with clustering can identify unusual patterns and groupings characteristic of coordinated fake accounts manipulating ratings.
Option a is incorrect because tokenization is a text processing technique that doesn't address the pattern detection needed for identifying coordinated fake accounts.
Option c is incorrect because grammar checking only validates language structure and cannot detect the behavioral patterns of Sybil attack fake identities.
Option d is incorrect because passive sensor monitoring observes without actively analyzing the patterns that reveal coordinated fake account activities.

Correct Answer: Dataset augmentation techniques

Option b is correct because dataset augmentation can generate training examples under various lighting conditions, making the model robust to illumination changes.
Option a is incorrect because passive sensing alone doesn't address the fundamental challenge of varying lighting conditions affecting defect detection accuracy.
Option c is incorrect because active sensing emits signals but doesn't specifically solve the problem of training models for varying lighting conditions.
Option d is incorrect because simple feature extraction without augmentation won't create the robustness needed for accurate detection under different illumination.

Correct Answer: Modal logic systems

Option b is correct because modal logic specifically deals with concepts of necessity and possibility, perfectly suited for reasoning about uncertainty levels.
Option a is incorrect because propositional logic handles true/false statements but cannot represent the nuanced degrees of certainty in human expressions.
Option c is incorrect because Boolean logic only handles binary true/false values without the capability to represent varying levels of certainty.
Option d is incorrect because linear temporal logic focuses on sequences over time rather than representing different degrees of certainty in statements.

Correct Answer: Semi-structured hybrid data

Option c is correct because semi-structured data combines organized sensor readings with flexible formats like traffic camera feeds for comprehensive pattern analysis.
Option a is incorrect because unstructured data alone lacks the organized temporal and spatial information essential for systematic traffic pattern prediction.
Option b is incorrect because structured sensor feeds alone miss important contextual information from cameras, weather data, and event information sources.
Option d is incorrect because quasi-structured log files like system logs don't capture the real-time traffic flow data needed for congestion prediction.

Correct Answer: Feature de-identification removal

Option c is correct because removing or de-identifying demographic features prevents the model from using biased correlations while preserving job-relevant qualifications.
Option a is incorrect because deleting all data would eliminate the training dataset entirely, making it impossible to build any hiring model.
Option b is incorrect because differential privacy adds noise for privacy protection but doesn't specifically address the demographic bias in hiring decisions.
Option d is incorrect because normalization scales numerical features but doesn't remove the underlying demographic bias present in the training data.

Correct Answer: Semi-supervised learning approach

Option b is correct because semi-supervised learning effectively combines small amounts of labeled purchase data with large volumes of unlabeled browsing patterns.
Option a is incorrect because purely supervised learning cannot utilize the valuable information contained in the unlabeled browsing data effectively.
Option c is incorrect because unsupervised clustering alone cannot leverage the labeled purchase history to predict specific buying behaviors accurately.
Option d is incorrect because reinforcement learning requires reward signals from actions rather than combining labeled and unlabeled customer data.

Correct Answer: Convolutional neural networks

Option b is correct because CNNs excel at processing grid-like image data and detecting spatial patterns like disease manifestations on crop surfaces.
Option a is incorrect because RNNs are designed for sequential data processing rather than identifying spatial patterns in agricultural imagery effectively.
Option c is incorrect because simple perceptrons lack the deep architecture and feature extraction capabilities needed for complex image pattern recognition.
Option d is incorrect because bidirectional RNNs process sequences in both directions but aren't optimized for spatial pattern detection in images.

Correct Answer: K-means clustering algorithm

Option b is correct because K-means clustering automatically groups similar articles based on content features without requiring predefined categories or labels.
Option a is incorrect because logistic regression is a supervised classification method that requires labeled training data with predefined categories to function.
Option c is incorrect because decision trees are primarily supervised learning methods that need labeled examples to build classification rules effectively.
Option d is incorrect because random forest is an ensemble of decision trees for supervised learning, requiring labeled data rather than unsupervised grouping.

Correct Answer: Differential privacy guarantees

Option b is correct because differential privacy provides mathematical guarantees that individual data points cannot be identified while maintaining statistical utility.
Option a is incorrect because simple anonymization can often be reversed through re-identification attacks and lacks mathematical privacy guarantees for individuals.
Option c is incorrect because encryption protects data in transit and storage but doesn't provide privacy guarantees during the model training process.
Option d is incorrect because access control lists manage who can view data but don't protect individual privacy during statistical analysis.

Correct Answer: Word embedding representations

Option b is correct because word embeddings capture semantic relationships, allowing the system to recognize that different phrases express similar intents.
Option a is incorrect because one-hot encoding creates sparse binary vectors that cannot capture semantic similarity between different expressions of the same intent.
Option c is incorrect because tokenization only breaks text into pieces without understanding semantic relationships between different ways of expressing commands.
Option d is incorrect because bag-of-words focuses on word frequency rather than semantic meaning, missing the relationships between varied expressions.

Correct Answer: Utility-based agent systems

Option b is correct because utility-based agents can evaluate multiple objectives and constraints to find optimal solutions balancing competing delivery requirements.
Option a is incorrect because simple reflex responses cannot handle the complex multi-objective optimization required for route planning with multiple constraints.
Option c is incorrect because passive learning only observes without actively optimizing for the multiple competing objectives in delivery route planning.
Option d is incorrect because single objective optimization cannot effectively balance the multiple competing constraints like time, distance, and weather conditions.

Correct Answer: Temporal data consistency

Option b is correct because consistent temporal data ensures equipment performance changes are accurately tracked over time for reliable failure prediction.
Option a is incorrect because while uniformity helps, it's less critical than maintaining consistent temporal sequences for detecting performance degradation patterns.
Option c is incorrect because basic relevance doesn't ensure the temporal consistency needed to detect subtle performance changes over time periods.
Option d is incorrect because veracity alone doesn't guarantee the temporal consistency required for tracking gradual equipment performance degradation patterns.