Ultimate Engineering Study Guide - Questions & Answers

True or False:All graphical models involve a number of parameters which isPOLYNOMIAL in the number of random variables.
1 Answer
Using the deterministic Model and given the following page reference string: 1,2,5,7,2,6,5,4,2,1,8,7,8,7,8,5,2,9,5,2,1,2,3,2,7,9. How many page faults would occur for each of the following 2 replacement algorithms assuming 4 frames? [Optimal, LRU] Use pure-demand paging. Show your work. LRU: OPT:
1 Answer
31. What's wrong with this model architecture: (6, 13, 1) a. the model has too many layers b. the model has too few layers C. the model should have the same or fewer nodes from one layer to the next d. nothing, looks ok 32. This method to prevent overfitting shrinks weights: a. dropout b. early stopping C. L1 or L2 regularization d. maxpooling 33. This method to prevent overfitting randomly sets weights to 0: a. dropout b. early stopping C. L1 or L2 regularization d. maxpooling 34. Which loss function would you choose for a multiclass classification problem? a. MSE b. MAE C. binary crossentropy d. categorical crossentropy 35. Select ALL that are true. Advantages of CNNs for image data include: a. CNN models are simpler than sequential models b. a pattern learned in one location will be recognized in other locations C. CNNs can learn hierarchical features in data d. none of the above 36. A convolution in CNN: a. happens with maxpooling. b. happens as a filter slides over data c. happens with pooling d. happens with the flatten operation 37. True or false. Maxpooling reduces the dimensions of the data. 38. True or false. LSTM suffers more from the vanishing gradient problem than an RNN 39. True or false. LSTM is simpler than GRU and trains faster. 40. True or false. Embeddings project count or index vectors to higher dimensional floating-point vectors. 41. True or false. The higher the embedding dimension, the less data required to learn the embeddings. 42. True or false. An n-dimensional embedding represents a word in n-dimensional space. 43. True or false. Embeddings are learned by a neural network focused on word context.
1 Answer
PART I We want to build a data warehouse to store information on country consultations. In particular, we want to know the number of consultations, in relation to different criteria (people, doctors, specialties, etc. This information is stored in the following relationships: PERSON (Person_id, name, phone, address, gender) DOCTOR (Dr_id, tel, address, specialty)CONSULTATION (Dr_id, Person_id, date, price) Tasks :1. What is the fact table? 2. What are the facts? 3. How many dimensions have been selected? What are they? 4. What are the dimension hierarchies? Draw them. 5. Propose a relational diagram that takes into account the date, the day of the week, month, quarter and year.
1 Answer
Prove that all regular languages can be recognized on be expressed usingA -> aBA->a a is terminal A, B are variables
1 Answer
Correlation between a factor (e.g. social support) and the ladder score (which presents happiness in this dataset).do countries that have a high ladder score generally have a high social support score?Does ladder score generally go up if social support score goes up?If so, is the correlation consistent across countries? If not, is it more significant in certain regions e.g. Europe but not the others?Consider using a scatter plot to explore the correlation. Also, please adjust the figure size so that all the labels are legible.I WAS usIng this program but I dont how to just and create a scatter plot to answer these questions world_happiness_report_2020.csvimport pandas as pdimport matplotlib.pyplot as pltdf = pd.read_csv('world_happiness_report_2020.csv')df.plot() # plots all columns against indexdf.plot(kind='scatter',x='Country name',y= 'Generosity') # scatter plotdf.plot(kind='density') # estimate density function# df.plot(kind='hist') # histogram
1 Answer
If the antivirus has a malware analyzer, what is the probability that a given malware will be detected in a 5000 mails as spam given that a spam is detected in the mail and the malware to spam detected ratio is 1/10.
1 Answer
1) Suppose we have Z = X * Y + W * Ua) Write the instruction with a three-address ISAb) Write the instruction with a two-address ISAc) Write the instruction with a one-address ISA
1 Answer
What is the point of the EM algorithm? Select the best option below. Be careful to consider the distinction between calculation of a probability (given some implicit parametric form) and maximization of a probability (by choosing the parameters directly.)A. The purpose of EM is to maximize the observed data likelihood P(X) when the joint likelihood P(X,Z) is tractable, but the hidden variables Z are not known. It does reduce the complexity of calculating P(X), so it works best when both P(X) and P(X,Z) can be evaluated in polynomial time.B. The purpose of EM is to maximize the observed data likelihood P(X) when the joint likelihood P(X,Z) is tractable, but the hidden variables Z are not known. It also allows us to tractably approximate the P(X) even when exact computation is exponential.C. The main application of EM is to obtain samples from the joint distribution P(X,Z) which can then be used as training data.D. EM can be used to handle exponential sums arising from inference problems. I.e., the EM algorithm canbe used to calculate P(X) in polynomial time even when there are many nusiance variables that have to be summed out from the joint distribution, P(X,Z).
1 Answer
True or False:Markov Chain Monte Carlo (MCMC) sampling algorithms work bysampling from a markov chain with a stationary distributionmatching the desired distribution.
1 Answer
17. This metric measures the percentage of items that were classified as + that were truly + TP/(TP + FP) a. precision b. recall C. accuracy d. F-measure 18. This metric is a balance of precision and recall. a. p-value b. accuracy C. F-measure d. none of the above 19. True or false. It is helpful to use a development set to tune parameters if we have a small amount of data. 20. True or false. Nave Bayes is a discriminative model. 21. True or false. Kappa ranges from 0 to 1. 22. True or false. The ideal AUC value is either +1 or -1. 23. This term refers to how well an algorithm can model different data sets. a. bias b. variance c. none of the above 24. Select ALL that are true. The purpose of adding a regularization term to an objective function is: a. to prevent underfitting b. to prevent overfitting c. to penalize large weights d. to penalize small weights 25. Select ALL that are true. Which are true about activation functions for neural networks: a. the sigmoid function output ranges from 0 to 1 b. the tanh function output ranges from -1 to +1 C. the rely output ranges from 0 to infinity d. the softmax function output sums to 1 26. True or false. Neural networks can have only one output 27. True or false. Logistic regression requires more feature engineering than neural networks. Deep Learning Questions 28. Trueor false. A layer represents a function that inputs tensors and outputs transformed tensors. 29. True or false. A model defines how neuro are put gether. 30. Select ALL that are true. Advantages of deep learning models over more shallow neural networks and traditional ML algorithms: a. they can learn more complex functions b. they can learn data representations at the same time as the function c. they train faster d. they require less data
1 Answer
9. Select ALL that are true. Nave Bayes a. typically has low bias b. typically has high bias c. can work well with small data sets d. performs poorly on small data sets P(A|B) = P(B|A) P(A) /P(B) 10. In the Bayes' Theorem formula above, the quantity P( AB) is a. called the posterior b. called the prior c. called the likelihood, or conditional probability d. used for normalization 11. In the Bayes' Theorem formula above, the quantity P(A) is a. called the posterior b. called the prior c. called the likelihood, or conditional probability d. used for normalization 12. In the Bayes' Theorem formula above, the quantity P(BIA) is a. called the posterior b. called the prior c. called the likelihood, or conditional probability d. used for normalization 13. In the Bayes' Theorem formula above, the quantity P(B) is a. called the posterior b. called the prior c. called the likelihood, or conditional probability d. used for normalization 14. True or false. Naive Bayes is a bag-of-words model. 15. This metric gives a percentage of correctly classified items of the total items classified. a. precision b. recall c. F-measure d. accuracy 16. This metric measures the percentage of items classified as + that were identified: TP/(TP + FN) a. precision b. recall c. F-measure d. accuracy
1 Answer
Lall-KAAs an Regular Expression and L(A) - ) Show that Lan is decidable.
1 Answer
Question 3 Given the two functions, f(n)= 2n+ 10 and g(n) = n, select the most suitable relationship between the two functions:O f(n) is in 2(g(n))O f(n) is in O(n) O f(n) is (g(n)) O f(n) is in o(g(n)) O f(n) is in O(g(n)) Question 4 Given the two growth functions, f(n) = n/100 + 10n - 100 and g(n) = 10n where n > 1, what is the smallest value of n (no) such that f(n) is in O(g(n))? O 100 O 20O 10 O 1000 O 11 Question 5 N is greater than 2. Select the tightest (best) lower bound of the growth rate, T(n) = n. O ohm(nlog(n)) O ohm(n/2) O ohm(log(n)) O ohm(n^0.5)O 22(n^0.9) Question 6 Suppose that a particular algorithm has a time complexity, T(n) = 8 * n/2 and a particular machine take t time for n inputs with this algorithm. If you are given a machine 216 times faster with the same algorithm. How many inputs could we process in the new machine in the same amount of time t? O n + 36 O n + 216 O 216n O n+6O 36n
1 Answer
Select each of the following states which are True (May be more than 1)1. Every directed graphical model can be converted to a NUMERICALLY equivalent undirected graphical model.2. All graphical models involve a number of parameters which is POLYNOMIAL in the number of random variables.3. Any UNDIRECTED graphical model can be converted into an DIRECTED graphical model with exactly the same STRUCTURAL independence relationships.4. When converting a directed graphical model to an undirected graphical model, the moralization process adds links between all pairs of co-parents (i.e., nodes which share a common child.)5. When converting a directed graphical model to an undirected graphical model, the moralization step adds links between all sibling nodes (i.e., between all pairs of nodes which share a common parent).6. Any probability distribution can be EXACTLY represented using an undirected graphical model.7. Any DIRECTED graphical model can be converted into an undirected graphical model with exactly the same STRUCTURAL independence relationships.
1 Answer
Is the following code segment valid although the identifier "three" is not typed?let three = 3var college = [Int]()college = [1,2,three]If yes, explain how. If not, suggest how to fix.In the above code segment, how to print the integer 3 from the array? Write a swift statement.In the above code segment, how to add the integer 4 to the array? Write a swift statement.
1 Answer
Select the statements which are TRUE below. (Correct one may more than one)1. The first and last observations are always conditionally independent of one another, given an intermediate observation.2. The first and last observations are always conditionally independent of one another, given an intermediate hidden state.3. The first and last hidden states are always conditionally independent, given an intermediate observation.4. The first and last hidden states are always conditionally independent, given an intermediate hidden state.
1 Answer
Summary:Considering a system with five processes PO through P4 and three resources of type A, B, C. Resource type A has10 instances, B has 5 instances and type C has 7 instances. Suppose at time tO following snapshot of the system hasbeen taken:Question1. What will be the content of the Need matrix? Question2. Is the system in a safe state? If Yes, then whatis the safe sequence?
1 Answer
Select the statements which are TRUE below. (Correct one may more than one)1. Markov Chain Monte Carlo (MCMC) sampling algorithms work by sampling from a markov chain with a stationary distribution matching the desired distribution.2. The Metropolis-Hastings algorithm (along with other MCMC algorithms) requires a period of burn-in at the beginning, during which time the initial configuration of random variables is adapted to match the stationary distribution.3. A significant advantage of MCMC algorithms (over, say, techniques such as rejection sampling) is that every iteration of the algorithm always generates a new independent sample from the target distribution.4. For MCMC to be "correct", the markov chain must be in a state of detailed balance with the target distribution.
1 Answer
a) Convert each of the following decimal values to 8-bit two's complement binary. i) -4810 ii) 6510 iii) -7510 iv) 8210
1 Answer
Find if the following system: y(n) = 5[x(n)]^2 + 10x(n) 1.Static or Dynamic 2. Causal or Non-Causal 3. Linear or Non-Linear4. Time Variant or Time Invariant 5. Stable or Unstable
1 Answer
1. In this type of machine learning, data is in the form (x, y) where x is a vector of predictor values and y is a target value, or label.* supervised learning* unsupervised learning* none of the above2. In this type of learning you do not have labeled data but are trying to find patterns in the data.* supervised learning* unsupervised learning* none of the above3. In this type of learning, you are building a model that can predict real-numbered values.O classificationO regressionO.both a and bO none of the above4. In this type of learning, your target is a finite set of possible discrete values.* classification* regression* both a and b* none of the above5. Select ALL that are true. Machine learning differs from traditional programming in that:* in ML, knowledge is not encoded in the algorithm (as in traditional programming)* ML programs learn from data* ML algorithms could get better over time* all of the above6. If your algorithm performs well on the training data but poorly on the test data, you have most likely:* underfit* overfit* neither7. What is the purpose of dividing data in train and test sets?O it gives us additional data on which to test the algorithmO it give us additional data to tune parametersO it allows us to give a more realistic evaluation of the algorithmO none of the above8. Nave Bayes is called nave because* it assumes that all predictors are dependent* it assumes that all predictors are independent* none of the above
1 Answer
5. Using a truth table to show that: a.x+x=1 for all values of x. b. y(x+x)=y for all values of x and y.
1 Answer
Question 1 Determine the result of the following arithmetic operations. (i) 3/2 (ii) 3.0/2 (iii) 3/2.0 Classify the type of statement for each of the following. (i) total=0; (ii) student++; (iii) System, out.println ("Pass"); Determine the output of the following statements. (i) System. out.println("1+2="+1+2); (ii) System.out.println("1+2=" +(1+2)); (iii) System.out.println(1+2+"abc"); Question 2 Explain the process of defining an array in the following line of code: int totalScore = new int [30];
1 Answer