Priya Dogra – Certification | Jobs | Internships
Related Articles
This course provides you with the fundamentals of the rapidly evolving field of artificial intelligence. Topics they will cover include the core AI concepts that are most relevant to today’s applications: Intelligent Agents, various kinds of machine learning models, search algorithms (including heuristic and uninformed search), different variations of iterative improvement algorithms, game playing, foundations of logic and automated reasoning, knowledge bases, natural language processing including generative AI, reasoning under uncertainty and how to apply these ideas to build intelligent software. You will need to know how to program in a modern language like Python, C#, or Java and how to apply libraries that are readily available from different vendors to apply the concepts you learn.
APPLY FOR THIS CERTIFICATION : CLICK HERE
Here are the exam answers :
Question 1 : Why is it difficult to precisely define intelligence?
Select one:
a.Intelligence is tied to human behavior
b.Intelligence has many complex elements
c.Intelligence is based on specific skills and brain power
Question 2 :What is the key difference between AI systems that “think like humans” vs AI systems that “act like humans”?
Select one:
a.AI systems that “think like humans” and AI systems that “act like humans” are fundamentally similar in design but perform differently at run time based on various conditions
b.AI systems that “think like humans” try to emulate processes in the brain whereas AI systems that “act like humans” perform functions that are thought to require intelligence
c.AI systems that “think like humans” perform functions that are thought to require intelligence whereas AI systems that “act like humans” try to emulate processes in the brain
Question 3 :How do AI solutions that “act rationally” become trustworthy?
Select one:
a.They produce results that people try and find useful and correct over time
b.They apply logical reasoning capabilities that usually yield the best outcomes
c.They usually yield good results because they emulate human brain processes
Question 4:How is a Learning Agent distinguished from every other type of agent?
Select one:
a. Learning Agent is able to improve its effectiveness based on continuous evaluation of its own key performance metricsb.
A Learning Agent is able to deduce its objectives from continuous evaluation of key aspects of its own problem environment
c.A Learning Agent is able to invent new algorithms and models from continuous analysis and evaluation of its own performance
Question 5 :The environment of chess is known to be fully observable and deterministic. Given this property, how should an agent be designed to play chess?
Select one:
a.The agent has to implement a strategy to look ahead on both its moves and its opponents’ moves
b.The agent has to simulate the entire game tree of moves for chess to ensure the opponent cannot win
c.The agent has to implement a strategy of sometimes making moves just to confuse the opponent
Question 6 :Why has machine learning assumed greater importance in today’s businesses?
Select one:
a.Businesses accumulate vast amounts of data that can be leveraged to find patterns to improve future outcomes
b.Businesses now have more financial resources to use advanced methods in their operations to deliver better outcomes
c.Businesses now have access to a huge pool of qualified software developers to write algorithms to improve outcomes
Question 7:How does reinforcement learning differ from other modes of machine learning?
Select one:
a.Reinforcement learning requires a training set that provides known outputs for given inputs
b.Reinforcement learning tries to organize data into clusters to make future classification possible
c.There is a value function that needs to be optimized, not a training set of inputs with known outputs
Question 8 :A robot is placed into a difficult environment to navigate for rewards without running into hazards or using up all its energy. What kind of ML model would be most applicable to this problem?
Select one:
a.Supervised learning
b.Unsupervised learning
c.Reinforcement learning
Question 9 :If a company wants to predict a person’s blood sugar level (a real number) from past data about sugar levels for persons given their age, weight, height, and various lifestyle choices, would you use a classification model or a linear regression model and why?
Select one:
a.Linear regression, because the sugar level can be treated as a label, and regression models can be used to predict the label
b.Linear regression, because the sugar level is a decimal number, and regression models can be used to predict numerical values
c.Classification, because the sugar level can be treated as a label, and classification leverages the training set to predict the label
Question 10:In the k-nearest neighbor machine learning model, what is the Manhattan distance between the following two points: A = {5, 2, 2 } and B = {1, 5, 2}
Select one:
a.6
b.7
c.8
Question 11 :A particular classification model is analyzed to exhibit high variance. What is the remedy?
Select one:
a.The classification model is too simple and needs to be enhanced to improve its prediction ability
b.The classification model is too complex and needs to be simplified to improve its prediction ability
c.The classification model is inappropriate and one needs to switch to a completely different algorithm
Question 12 :Using the linear regression algorithm on this data, what is the predicted value of Y for X=5?
X=3, Y=5
X=4, Y=9
X=7, Y=12
Select one:
a.Y = 8.19
b.Y = 9.19
c.Y = 10.19
Question 13:What is the significance of squaring the distance values in linear regression?
Select one:
a.This improves the accuracy by simply maximizing all the computed error values
b.This improves the accuracy of metrics because negative and positive errors will not cancel each other out
c.This improves the run-time efficiency of the algorithm by making sure all the distance values count more
Question 14 :When does unsupervised learning become useful?
Select one:
a.When you have complex datasets with well-understood models
b.When you have datasets with little understanding of what is in them
c.When you have complex datasets with some known inputs and outputs
Question 15 : Why is a Markov Decision Process useful as a simple model for Reinforcement Learning?
Select one:
a.The future is randomly chosen irrespective of the past states and present state
b.The future depends on the past states, present state, and the action chosen in the present state
c.The future depends only on the present state and action chosen, but not how you got to the present state
Question 16 : Which statement about Neural Networks (NNs) is true?
Select one:
a.NNs do not assume a known function between inputs and outputs
b.NNs do not usually have many internal layers that need to be trained
c.NNs can be trained very quickly, but may take more time to predict outputs
Question 17 :A healthcare screening company has accumulated millions of anonymous patient records with medical data and whether the patient was screened as a heart-attack risk or not. You are asked to use this past data to determine if new patients are either at risk for heart disease or Normal. What are the key ML requirements in this problem?
Select one:
a.Unsupervised learning: use standard statistics to predict the label for a new patient
b.Supervised learning: train classification models on existing data to predict the label for a new patient
c.Supervised learning: train regression models on the existing data to predict the label for a new patient
Question 18:In applying any supervised machine learning model with a large training dataset, you want to ensure the model is trained in the best way possible based on the training dataset. How much of the available dataset should you select randomly to train the model, and how much to validate the model?
Select one:
a.30% training, 70% validation
b.50% training, 50% validation
c.80% training, 20% validation
Question 19:What causes the combinatorial explosion in the complexity of a general problem solver (GPS)?
Select one:
a.The goal state is usually not well-defined
b.The operators are generally unpredictable in their effects
c.The number of operators that apply in each state is potentially very large
Question 20:When formulating a simple problem search tree for a general problem solver (GPS) one usually assumes that the environment is static. Why is this important?
Select one:
a.GPS assumes that the state of the environment can never be changed at any time by any means
b.GPS assumes that the state of the environment is not going to change when the GPS is deliberating
c.GPS assumes that the state of the environment changes in known ways when the GPS is deliberating
Question 21 :The environment for a self-driving car is considered dynamic. How does this property complicate the design of the car?
Select one:
a.The car must make decisions based on both the future and prior states, which means it has to look ahead
b.The car must make decisions fast and validate actions prior to implementation to ensure they are still applicable
c.The car must make decisions assuming that the actions are always valid when they are ready to be implemented
Question 22:How does a general problem solver (GPS) find a solution to a problem?
Select one:
a.From the initial state (root), expand the tree by level using operators until the goal state is reached
b.Starting from the initial state (root), randomly enumerate all paths to the goal state using operators
c.From the goal state, work backwards to previous states using reverse operators (if possible) to reach the initial state (root)
Question 23:All uninformed search algorithms can be viewed as tree expansion with different queuing schemes. Which of the uninformed search algorithms uses a Last In First Out queuing scheme?
Select one:
a.Depth First Search
b.Breadth First Search
c.Uniform Cost Search
Question 24 :If you apply depth-first search to this search tree, with an open queue and an expanded queue as shown, what will be the next node to be expanded? (When there is a tie, select the node that is lower in alphabetic order. If there is a choice between W, Y, and Z, the answer will be W because it is lower in alphabetic order.)
Select one:
a.E
b.G
c.H
Question 25 : What is an admissible heuristic in the context of search algorithms?
Select one:
a.The heuristic estimate of the cost for any node to reach a goal state cannot exceed the actual cost for that node
b.The heuristic estimate of the cost for any node to reach a goal state must exceed the actual cost for that node
c.The heuristic estimate of the cost for any node to reach a goal state is identical to the actual cost for that node
Question 26 :If you apply an A* search to this search tree, using admissible heuristic h2, with an open queue and an expanded queue as shown, what will be the next node to be expanded? (When there is a tie, select the node that is lower in alphabetic order. If there is a choice between W, Y, and Z, the answer will be W because it is lower in alphabetic order.)
search tree
Select one:
a.E
b.G
c.H
Question 27:In which kind of problems can iterative improvement algorithms be useful?
Select one:
a.he goal state is completely irrelevant, and the path is the solution
b.The goal state cannot be described, whereas the path can be described
c.The goal state is itself the solution and the path to get the solution is irrelevant
Question 28:How would you apply the iterative improvement concept to the N-Queens problem (Placing N queens on an N × N chessboard, N >= 4, such that no queen attacks any other queen)?
Select one:
a.Place the N queens on the board anywhere; move each queen to a position where it is not being attacked; back-track if stuck to the previous step; stop when ALL queens are on the board and non-attacked
b.Place the first queen on the board; add the next queen so that it is not being attacked by the queens on the board; repeat until all queens are placed; if stuck, there is no solution to this problem
c.Place the first queen on the board; add the next queen so that it is not being attacked by the queens on the board; repeat until all queens are placed; if stuck, repeat the process from a different first position
Question 29:Why is the hill-climbing algorithm not guaranteed to find a global optimum?
Select one:
a.The algorithm will never reach either a local or global optimum
b.The algorithm is designed to always move quickly away from local optima
c.Once the algorithm reaches a local optimum, it will not accept any new moves
Question 30:What is the key difference between applying the standard iterative improvement algorithm and the simulated annealing algorithm to the traveling salesman problem (finding the best route for a salesman starting from City A, ending at City X, and visiting all the cities where she has customers)?
Select one:
a.Simulated annealing will always accept a sub-optimal route swap, whereas iterative improvement will always reject that route swap
b.Simulated annealing will accept a sub-optimal route swap with some probability, whereas iterative improvement will always reject that route swap
c.Simulated annealing will never accept a sub-optimal route swap, whereas iterative improvement will always accept that route swap with some probability
Question 31 : How does a genetic algorithm work in terms of iterative improvement?
Select one:
a.It incorporates mechanisms to gradually explore the search space by using heuristic search
b.It incorporates mechanisms to combine elements of 2 bad solutions to create better solutions
c.It incorporates mechanisms to gradually explore the search space by using an uninformed search
Question 32:What is the normal assumption underlying any representation of two-person games?
Select one:
a.That the players may make arbitrary decisions at every stage based on how they feel
b.That the players will always make the rationally correct and best possible move at every stage
c.That the players, being human, may make mistakes and that the tree should represent that possibility
Question 33:If the Minimax algorithm is implemented, which statement about how Min and Max will play the game is true?
Select one:
a.Max and Min are both free to select a successor with any value randomly
b.Max will select a successor with the highest value, and Min will select a successor with the lowest value
c.Max will select a successor with the lowest value, and Min will select a successor with the highest value
Question 34:What is the reason that alpha-beta pruning can improve upon the Minimax algorithm?
Select one:
a.Alpha-beta pruning alternatively ignores Max or Min nodes to improve its speed
b.Alpha-beta pruning will never explore any node whose value cannot impact the decision of its parent
c.Alpha-beta pruning randomly ignores nodes, assuming that they cannot impact the decision of their parents
Question 35:What is the important architectural property of a peer-to-peer model for multi-person games?
Select one:
a.The model supports no communication between the players, but they can guess the moves
b.The model supports communication between a player and other players without coordination
c.The model supports communication between a player and other players through central coordination
Question 36 : Which of these business applications leverages the power of natural language processing?
Select one:
a.A user support chatbot that accepts questions and tries to answer them automatically
b.A video that walks a user, screen by screen, through the steps to configure an appliance
c.A data entry form where users to describe their problems in English and submit them to support staff
Question 37:What is a confounding characteristic of NL that makes it difficult to process automatically?
Select one:
a.NL contains idiomatic expressions that vary greatly between different contexts and cultures
b.NL is inherently ambiguous because there are many interpretations of the same text based on context
c.NL often involves implicit meanings that require a deep understanding of the surrounding context to interpret correctly
Question 38:How does named entity recognition from a corpus of text work?
Select one:
a.Look up all the capitalized letters in an entity database, just like a keyword search
b.Apply natural language generation to generate named entities and check against the corpus
c.Use NLP to find syntactic/semantic phrases and look up plausible phrases in an entity database
Question 39 : Why is NLP essential to sentiment analysis?
Select one:
a.NLP provides models of how the brain works, applies them to the training text samples, and generalizes the models to predict sentiments of new text
b.NLP breaks training text down into semantically significant blocks, associates those components with the known sentiment, and applies machine learning to compute sentiments of new text
c.NLP gets all the way down to the bit-level representation of training text, associates them with the known sentiments, and applies machine learning to compute sentiments of new text represented as bits
Question 40 : Why is LSTM useful in voice recognition applications?
Select one:
a.LSTM detects long sequential speech patterns and responds effectively
b.LSTM fares better with sparse input that comes from applications like Voice Recognition
c.LSTM provides faster response in generating responses to detected patterns in spoken speech
Question 41: How does statistical NLP function?
Select one:
a.It accumulates and analyzes syntactic and semantic data from NLP analysis
b.It uses the syntactic and semantic data from NLP analysis to draw inferences
c.It views the text as noisy data and quantitatively analyzes it for various signals to make predictions
Question 42 :How does Natural Language Generation (NLG) create new content?
Select one:
a.It uses statistical NLP to generate the appropriate outputs for any new inputs
b.It uses search mechanisms to find content based on the inputs and shows that to the user
c.It uses ML models trained on many inputs with known outputs to generate the right output for new input
Question 43:What is the primary difference between a long short-term memory model (LSTM) and a large language model (LLM)?
Select one:
a.LSTMs are based on recurrent neural networks, and LLMs use convoluted neural networks, which are faster
b.LSTMs are based on recurrent neural networks, whereas LLMs use transformers to process text into new text rapidly
c.LSTMs are based on convolutional neural networks, whereas LLMs use transformers to process text into new text rapidly
Question 44:Which situation would be most difficult to describe in propositional logic (PL)?
Select one:
a.It is raining
b.It rains every Thursday
c.Today is Thursday and it is raining
Question 45:The modus ponens inference procedure goes hand-in-hand with knowledge bases that are in Horn Form. When would you use modus ponens with backward chaining versus forward chaining?
Select one:
a.Backward chaining and forward chaining both solve the same inference problems, but forward chaining performs better in most situations with limited memory
b.Backward chaining works best to prove a specific sentence, whereas forward chaining works best in enumerating all the sentences entailed by a knowledge base
c.Forward chaining works best to prove a specific sentence, whereas backward chaining works best in enumerating all the sentences entailed by a knowledge base
Question 46 :How does first-order logic (FOL) differ from propositional logic (PL)?
Select one:
a.FOL removes logical connectives from PL
b.FOL adds objects, predicates, relationships, functions, and quantifiers to PL
c.FOL provides the capability to write more sentences using the same capabilities as PL
Question 47:Why is it important for an FOL knowledge base (KB) to be expressed in Conjunctive Normal Form (CNF) before we can apply resolution as an inference procedure?
Select one:
a.The KB expressed in CNF omits many clauses automatically, thereby considerably simplifying the inference procedure
b.The KB expressed in CNF simplifies many clauses by rewriting them, making the resolution procedure much easier to apply
c.The KB expressed in CNF allows for more efficient automation of resolution by eliminating a clause and its negation in each iteration
Question 48 :Why does probability theory provide a solid foundation for reasoning under uncertainty?
Select one:
a.The probability of an event represents the likelihood of that event occurring without modeling all the details of the event
b.The probability of an event provides important causal details about the factors that contribute to the occurring of that event
c.The probability of an event provides a logical model to describe all the needed details of what can cause the event to occur
Question 49:Let C be Covid (a disease), and let HT be “High Temperature”, a symptom of various diseases, including Covid. Further, from existing statistics, say that we know the following:
Probability that a patient with Covid has High Temperature: P(C|HT): 0.9
Probability of Covid, P(C), in the population is: 0.00001
Probability that patients have High Temperature, P(HT) in the population is: 0.08
Given the information above, what is the probability that a person with high temperature as a symptom has Covid (P(C|HT))?
Select one:
a.:0.0001125
b.0.001125
c.0.01125
Question 50 :This figure shows the dependencies between three events: burglary (B), earthquake (E), and alarm (A). The probabilities associated with independent events B and E are shown. According to this conditional probability table, the alarm event (A) is dependent on B and E.
the dependencies between three events: burglary (B), Earthquake (E), and Alarm (A)
What is the probability that there is a burglary (B), no earthquake (E), and no alarm (A) – in mathematical terms, what is P (B⋏¬E⋏¬A)?
Select one:
a.0.00000598
b.0.0000598
c.0.000598
Question 51:What is the value of Markov chains in modeling uncertainty in systems?
Select one:
a.Any number of successors can be specified to each state with strict rules about transitions
b.Any number of successors can be specified to each state, and transitions happen randomly
c.Transition probabilities from a state to each of its successors describe various possible outcomes
