ICS 1020 Mock Questions PDF

ICS 1020 Mock questions: 1. What are the four main areas of Artificial Intelligence? A. Machine Learning, Reinforcement Learning, Problem Solving and Vision B. Natural Language Processing, Vision Systems, Game Theory and Robotics C. Reasoning, Learning, Problem Solving and Perception...

ICS 1020 Mock questions: 1. What are the four main areas of Artificial Intelligence? A. Machine Learning, Reinforcement Learning, Problem Solving and Vision B. Natural Language Processing, Vision Systems, Game Theory and Robotics C. Reasoning, Learning, Problem Solving and Perception D. Autonomous Vehicles, Conversational AI, Games and Search. 2. How do we distinguish a discrete environment from a continuous one? A. A discrete environment is one which has unobservable information, while a continuous one is one which can be continuously observed without interruption. B. A discrete environment is one with a finite number of action choices and states, while a continuous one has an infinite number of possible states or actions. C. A discrete environment is one which deals with real-valued numbers in the environment states, while a continuous one is one does not. D. A discrete environment is one which has clear termination criteria, such as the end of a game, while a continuous environment is one which does not terminate, such as the stock market or controlling the transport infrastructure. 3. Which of the following is an adversarial problem setting? A. An intelligent agent playing a chess game. B. A system trying to predict the weather. C. A Rubik’s cube solver. D. A system which detects objects from images. 4. What is the difference between blind-search and informed-search algorithms? A. Blind-search algorithms do not need to generate all the search space, while informed-search algorithms generate all the search-space before starting the search. B. Blind-search algorithms search systematically all the nodes in a predefined order until a solution is found, while informed-search algorithms use an evaluation function to determine which node is best to expand next. C. Blind-search algorithms do not look ahead into the successors of each node, while informed-search algorithms look one step ahead. D. Blind-search algorithms use a heuristic function to guide them through the search space, while informed-search algorithms do not need this because they already know which nodes are best to expand. 5. Which of these statements is not true about Local (Neighbourhood) Search algorithms? A. They are typically used for optimization problems. B. They are useful to search through very large, possibly infinite or continuous search spaces. C. They guarantee to find the globally optimal solution. D. They work by improving the current solution incrementally with small changes. 6. What is a Knowledge Base? A. A database of facts extracted from a reliable internet source such as Wikipedia. B. A repository of information which is indexed in a way such that it can be queried using natural language. C. A repository of statements, known as sentences, which can either be axioms, or inferred from other sentences using inference rules. D. Any body of knowledge about a specific domain for which an intelligent agent is going to be used. 7. What is a term in First Order Logic? A. A term refers to an object that could be a constant, a function of another object, or a variable that acts as a placeholder for any object that fits some criteria. B. A term is a sentence made up of atoms and connectives, such as negation, conjunction and disjunction. C. A term is an atom or a negated atom. D. A term is a clause part of a conjunction or a disjunction. 8. What is the difference between Domain-Specific planning and Domain-Independent Planning? A. Domain-specific planning is flexible to work across multiple domains, while domain-independent planning does not allow for such flexibility. B. Domain-specific planning is faster than domain-independent planning. C. Domain-specific planning requires less memory than domain-independent planning. D. Domain-specific planning systems are designed to work for only one application, while domain-independent planning systems apply generic algorithms that are independent of the problem being solved. 9. An A.I. Planning task is often solved by encoding it as a state-space search problem. How is this encoding done? A. Nodes correspond to environment states, while edges correspond to applicable actions that lead to other states. The edges on the path from the initial state node to a goal node corresponds to the actions in a valid plan. B. Nodes correspond to possible actions while edges represent the effect of the actions on the environment. The nodes in the path that leads to an edge that has the goal in its effect constitute a plan. C. States are organized in a tree structure, with the initial state at the top, and expanded until a state that corresponds to a goal is found. D. Each node corresponds to plan, and each edge is a modification that leads to another plan. A valid plan is found when a node that satisfies the goal is found. 10. Given a fair 6-sided dice, what is the probability of throwing the dice 2 times and getting a total of 6? A. 1/6 B. 1/12 C. 1/24 D. 1/36 E. 5/36 11. What do we mean by Independence in probability theory? A. An event, A, is independent of an event B, if the probability of A is influenced by the outcome of B. B. An event, A, is independent of an event B, if the probability of A is not influenced by the outcome of B, and vice versa. C. An event, A, is independent of an event B, if the probability of A is influenced by the outcome of B, given another known outcome C. D. An event A, is independent of an event B, if the outcome of A influences the probability of B. 12. What do we mean by overfitting in Machine Learning? A. When the accuracy of a classification algorithm is too good to be true. B. When the machine learning algorithm is left to train for too long and it starts to lose its predictive performance. C. When a machine learning algorithm is supplied to much data to train. D. When a machine learning algorithm performs well on the training data but fails to generalise and performs badly on test data. 13. Which of the following is typically not implemented using an unsupervised learning algorithm? A. Clustering B. Anomaly Detection C. Regression D. Dimensionality Reduction 14. What is a Policy in the context of Reinforcement Learning? A. A set of rules that the reinforcement learning agent must follow. B. A function that given a state tells the intelligent agent what action to take. C. A graph of nodes and edges with probabilities and rewards. D. A sequence of actions that must be executed in order by the agent. 15. What is a Markov Decision Process? A. A directed graph with state nodes and action nodes, and edges that indicate which actions are applicable in each state, and what is the probability of reaching a subsequent state, together with any reward. B. A graph of nodes and edges, where nodes represent states and edges represent actions. C. A tree of nodes and edges, where each layer of nodes represents the decision points of a different agent. D. A process through which an intelligent agent decides which actions to take. 16. Which of the following is not considered to be a zero-sum game? A. Stock trading where participants buy and sell company shares and financial assets to each other. B. A game of chess, where the winner gets 1 point and the loser gets a negative 1 point. C. A football league where for each match the winning team gets 3 points and in case of a draw each team gets 1 point. The losing team gets 0 points. D. A game of poker, where the winner of each round gains the money lost by all the opponents in that round. 17. What is a mixed strategy? A. A strategy where the player chooses different strategies to confuse the opponent. B. A situation where the player does not know which is the best strategy or is confused about what to do. C. A situation where the strategy is chosen randomly with some predefined probability. D. A strategy which does not choose the most rational decision. 18. What are the main issues we have to consider when dealing with data perceived from sensors? A. The reliability of the sensors, and how can we verify the data we are reading. B. What unobservable information do we have to take into consideration, and is there a way to infer it. C. How long should the perceived data be considered valid for. D. All of the above. 19. A flying drone can move up and down, left and right, forward and backward, and change orientation in 3 directions (roll, pitch and yaw). How many variables do we need to keep to represent the dynamic state of the drone? A. 6 B. 9 C. 12 D. 15 20. Object detection A. Has evolved over the past 20 years to include deep learning models B. Is used mostly in the film industry C. Uses only 2 important algorithms, which operate in a one stage detection method D. That uses the one stage detection method gives much more accurate results 21. Computer Vision applications in the Healthcare Industry: A. Are being used successfully in areas such as medical imagery, diagnostic applications as well as screening. B. Are mostly used to assist in surgeries. C. Are limited in scope because of patient privacy. D. Are used to inspect the quality of prescription drugs. 22. What is a stop word in NLP? A. A word that is commonly used in a language but carries little meaning. B. A word that is used to connect other words in a sentence. C. A word that is commonly used in a language but is difficult for a computer to understand. D. A word that is used to improve the accuracy of language recognition software. 23. What is tokenization in NLP? A. The process of breaking down a text into its component words or phrases. B. The process of assigning a numerical value to each word in a text. C. The process of grouping words together based on their meaning. D. The process of removing stop words from a text. 24. On the point of automation, ethically: A. AI is liable for any loss of jobs in the field of work. B. Care has to be taken so that the AI does not get out of control. C. Guidelines need to be set to protect humans against possible discrimination. D. AI is not concerned with human interaction. 25. Algorithmic Fairness: A. Is a field of machine learning that deals with mitigating possible instances of data bias. B. Is concerned with making the algorithm as fair as possible. C. Includes ways in which the data is changed to make the results as fair as possible to everyone. D. Results in the misalignment of causal complex relationships to make way for more accurate results.