STUDENT ID NO MULTIMEDIA e UNIVERSITY | MULTIMEDIA UNIVERSITY FINAL EXAMINATION TRIMESTER 2, 2018/2019 TPR2251 — PATTERN RECOGNITION (All sections / Groups ) 01 MARCH 2019 03:00 p.m. ~ 05:00 p.m. (2 Hours ) INSTRUCTIONS TO STUDENTS 1. This question paper consists of 5 pages with 5 questions only. 2. Attempt ALL questions, All questions carry equal marks and the distribution of the marks for each question is given. 3, Please print all your answers in the answer booklet provided.era2s1 PATTERN RECOGNITION 01 MARCH 2019 QUESTION 1 (@ Car plate recognition is a popular application of pattern recognition techniques. Explain the FOUR important processes in a car plate recognition system. {4 marks} (b) Suppose you have designed two classifiers shown in Figure 1 to predict whether it will rain’ the next day based on today’s temperature and humidity measures. Which model (Model 1 or Model 2) will be more appropriate for the weather prediction problem? Justify your answer Model 1 Model 2 Figure 1. Two classifiers (Model I and 2) for weather prediction. [2 marks} (© A genetic test has a probability 0.96 of giving a positive result when applied to a person suffering from a certain genetic defect, and a probability 0.08 of giving a ————false-positive-when applied to a non-sufferer. It is estimated that 0.2% of the population suffers from the genetic defect. Suppose that the test is now administered to a person. Calculate the following probabilities that: (i) the test result will be positive. [I mark) (ii) given a positive result, the person suffers from the defect. {I mark) (ii) given a negative result, the person is a non-sufferer. [1 mark) (iv) the person will be misclassified. {1 marlq, Continued ....... Levaren 1/5Teraast PATTERN RECOGNITION ob MARCH 2019 QUESTION 2 (a) __ Given the following two-class pattern recognition problem: You are designing a classifier to separate two kinds of fish: tenggiri and cencaru. You need to predict what type of fish will emerge next on the conveyer belt and that the sequence of types of fish appears to be random. (i) What are prior probabilities in the problem? [I mark] Gi) Name FOUR features that can be used in the classifier. [2 marks] (b) _Inatwo-class problem with a single feature x, the likelihood functions for «, and 2 are given as follows: inimum error classification and the [3 marks] If P(@,) = 2/3 and P(w2) = 1/3, find the corresponding expected classification error for (©) _ Ina three-class, two-dimensional problem, the feature vectors in each class are normally distributed with a covariance matrix 1.2 04 rales te The mean vectors for each class are [0.1 0.1)", (2.1 1.9]?, and(-15 2.0)", respectively, Assuming that the classes are equiprobable, classify the feature vector [1.6 1.5]" according to the Bayes minimum error probability classifier. [4 marks] Continued . Levan a5pR2251 PATTERN RECOGNITION 01 MARCH 2019 QUESTION 3 (@) What is dimension reduction? Explain why dimension reduction is crucial in a pattern recognition system like face recognition. (3 marks} () In your opinion, how is &-NN different from é-means clustering? Explain your answer. (3 marks] (©) Suppose you are building a classifier using Principal Component Analysis (PCA) to predict landmark buildings in Melaka. You ere given some input training images as shown in Figure 2. What pre-processing steps are required on these images if you want to get a reasonably good performance from the PCA algorithm? List TWO steps. (2 marks} ‘A Famosa Christ Church Figure 2. Three landmark buildings in Melaka, (4) Given a snapshot showing the scatter plot of two features, X1 and X2, comprising of two classes. Do you think Principal Component Analysis (PCA) can produce a ‘g00d classification result on this data? Justify your answer. Figure 3. Scatter plot for two features, X1 and X2. [2 marks} Continued ....... Levaren 375TeRD281 PATTERN RECOGNITION QUESTION 4 (1 MARCH 2019 (@) What is the difference between a generative and discriminative model? Explain. [2 marks] (b) You want to cluster the following cight points into three clusters using k-means algorithm: A, (2,10), A (2,5), A3(8,4), 44(5,8), 45(7,5), 4g(6,4), Ay(1,2), and Ag(4,9). The distance function between two points, @ = (x,,)4) and b= @2,¥2), is given by d(a,b) = [x2 — x4) + lye — yal. (@ Suppose the initial cluster centers are A,, Ay, and A>. Find the cluster centers after the first iteration, (ii) The cluster centers do not change anymore after the forth iteration. Predict the class of a new data point X(6,10) if the new cluster centers are [4 marks] (2,4), C2(3,6), and , C3 (7,9), respectively, [2 marks} (©) Given the following six points with their two corresponding attributes: Points ‘Attribute 7 ‘Aaribute 2 % 0.4005 0.5306 x; 0.2148 0.3854 xX 0.3457 0.3156 Xe 0.2652 0.1875 Xs 0.0789 04139 x 0.4548 0.3022 ‘The pair-wise distance matrix for the points are given by, % x x Xe X Xs x 0.0000 | 0.2357 | 0.2218 | 03688 | 03421 | 0.2347 x 0.2357 | 0.0000 | 0.1483 [0.2042 | 0.1388 | 0.2540 x 0.2218 [0.1483 [0.0000 [0.1513 | 0.2843 | 0.1100 x4 | 0.3688 | 0.2042 | 0.1513 | 0,0000 | 0.2932 | 0.2016 Xs 0.3421 [0.1388 [0.2843 | 0.2932 | 0.0000 | 0.3921 xs | 0.2347 | 0.2540 | 0.1100 | 0.2216 | 0.3921 | 0.0000 Plot the dendogram implemented using complete linkege proximity function. [2 marks] Continued .... LcvaTeN a5srpRa2st PATTERN RECOGNITION, 01 MARCH 2019 QUESTION 5 (@) With a supervised leaming algorithm, we can specify target output values but we may never get close to those targets at the end of leaming, Give TWO reasons why this might happen. [2 marks] (b) Consider a SVM classifier with a polynomial kernel of degree two and answer the questions below. @ Suppose you applied this model to your data and found that it perfectly fits the data (e.g., both training and testing accuracies are 100%). What would you observe if the degree of the polynomial kernel is increased from two to five? {1 mark} Gi) After increasing the complexity of the model, you found that the training accuracy was still 100%. What are the reasons behind this? Give TWO reasons, [2 marks} (©) What is a training set and how is it used to train neural networks? [2 marks] (@) Logical operators (i.e. NOT, AND, OR, and XOR) are the building blocks of ny computational device, Logical functions return only two possible values, true or false, based on the truth or false values of their arguments. Verify that the AND logical function can be implemented using the following single-unit with inputs: Am y= ev) where w, = 1 and w, = 1 represent the weights, respectively, and v is the threshold level. The activation is given by, 1 ifve2 9 = (9 ctherwise Hint: You may denote the truth value by J and false by 0. [3 marks} End of Page ievaren 375