Artificial Intelligence In Power System Optimization 1st Edition Weerakorn Ongsakul, Vo Ngoc Dieu - Solutions

Looking to master the complexities of "Artificial Intelligence In Power System Optimization" by Weerakorn Ongsakul and Vo Ngoc Dieu? Our platform offers comprehensive resources to guide you through each chapter. Access the online answers key and solutions pdf for every solved problem. Dive into the solution manual and instructor manual for detailed, step-by-step answers to all questions and answers provided in the textbook. Whether you're preparing for exams or need a test bank for practice, our chapter solutions are designed for your success. Discover the ease of learning with our free download options today.

=+ Strengths and weaknesses of these saliency
=+vanilla backpropagation, guided backpropagation, guided Grad-CAM, and SmoothGrad
=+using saliency mapping techniques, such as
=+ How to interpret convolutional neural networks
=+ How to implement convolutional neural networks for image classification tasks
=+ Characteristics that make convolutional neural networks inherently black-box
=+ Strengths and weaknesses of LIME, SHAP, and anchors
=+ How to interpret deep neural networks using LIME, SHAP, and anchors
=+are local in scope, such as LIME, SHAP and anchors
=+ Perturbation-based model-agnostic methods that
=+ How to implement deep neural networks that are inherently black-box models
=+Characteristics of deep neural networks
=+ How to uncover bias by looking at feature interactions
=+ How to interpret black-box models using a modelagnostic method called partial dependence plots (PDPs)
=+ How to interpret random forest models
=+ How to implement tree ensembles, specifically random forest—a black-box model
=+Characteristics of model-agnostic methods and global interpretability
=+ What black-box models are and their characteristics that make them inherently opaque
=+ How to implement and interpret GAMs
=+ What generalized additive models (GAMs) are and their properties that give them high predictive power and high interpretability
=+ How to interpret simple white-box models such as linear regression and decision trees
=+inherently transparent and interpretable
=+Characteristics that make white-box models
10 What is data mining? What is the difference between a query tool and data mining?What are data mining tools? How is data mining applied in practice? Give examples.
9 Why do we need to massage data before using them in a neural network model? How do we massage the data? Give examples of massaging continuous and discrete data.What is 1 of N coding?
8 What is the basis for the popularity of neural networks? What are the most successful areas of neural network applications? Explain why and give examples.
7 Why are fuzzy systems particularly well suited for modelling human decision-making?Why does fuzzy technology have great potential in such areas as business and finance?
12 What are the main steps in genetic programming? Draw a flowchart that implements these steps. What are advantages of genetic programming?
10 What is genetic programming? How does it work? Why has LISP become the main language for genetic programming?
11 What are the main steps in developing a frame-based expert system?
10 What are the differences, if any, between rules used in rule-based expert systems and those used in frame-based systems?
3 What is a facet? Give examples of various types of facets.
11 What are the differences between Mamdani-type and Sugeno-type fuzzy inferences?What is a singleton?
12 List advantages of rule-based expert systems. What are their disadvantages?
11 What is a conflict set of rules? How can we resolve a conflict? List and describe the basic conflict resolution methods.
7 Can an expert system make mistakes? Why?
12 What are the benefits of integrating expert systems, fuzzy logic and neural computing?
11 What are the main advantages of applying fuzzy logic in knowledge-based systems?
10 What are the premises on which fuzzy logic is based? When was fuzzy set theory introduced?
9 Why was the field of ANN reborn in the 1980s?
7 What are the limitations of expert systems?
21.3 Run MINIHYPER with the predecessor exercise and measure the execution time. How many hypotheses are generated, how many are refined? The 'guard' literal atom(X) in the background literals above is helpful, but there is a problem. At the moment that a literal pair like [atom(X),parent(X,Y)] is
21.2 Work out how many refinement steps are needed to obtain the target hypothesis from the start hypothesis above.
21.1 Experiment with MINIHYPER with modified sets of examples about has_daughter. How do these modifications affect the results?
20.6 Write a procedure prunetree( Tree, PrunedTree) that prunes a decision tree Tree according to the minimum-error pruning method discussed in this section. The leaves of Tree contain class frequency information represented as lists of integers. Use the Laplace estimate in the program. Let the
20.5 A volcano ejects lava completely randomly. In long-term statistics, the volcano was, on average, active one day in ten days and inactive the remaining days. In the most recent 30 days it was active 25 days. This recently increased activity was taken into account by experts in the forecast for
20.4 Complete the tree induction program of this section and test it on some learning problem, for example the one in Figure 20.9. Note that the procedure choose_attribute, using setof, is very inefficient and can be improved. Also add a procedure show( Decision Tree) for displaying decision trees
20.3 Implement a chosen impurity measure by writing the impurity1 predicate. This measure can be, for example, the residual information content or Gini index as discussed previously in this section. For the examples in Figure 20.9 and attribute size, use of the Gini index as impurity measure gives:
20.2 Disease D occurs in 25 percent of all the cases. Symptom S is observed in 75 percent of the patients suffering from disease D, and only in one sixth of other cases. Suppose we are building a decision tree for diagnosing disease D, so the classes are just D (person has D) and ~D (does not have
20.1 Consider the problem of learning from objects' silhouettes (Figure 20.9). Calculate the entropy of the whole example set with respect to class, the residual information for attributes 'size' and 'holes', the corresponding information gains and gain ratios. Estimate the probabilities needed in
19.3 Experiment with the program in Figure 19.4. Define other circuits. For example, extend the circuit of Figure 19.5(b) by adding a diode in series with resistor R5. How does this affect the voltage at T51? Try also the opposite orientation of the diode.
19.2 The program in Figure 19.2 initializes the upper bound on the finishing times (bestsofar) to a high value that is obviously a gross overestimate. This makes sure that an optimal schedule is within the bound, and will be found by the program. This policy, although safe, is inefficient because
19.1 Experiment with the program in Figure 19.2. Try different resource specifications, aiming at removing useless permutations, and timing the program's running times. What are the improvements?
18.5 Experiment with the POP planner and GRAPHPLAN program of this chapter in various planning domains (like robots on grid, blocks world, Sokoban), and compare the solutions and efficiency of both planners.
18.4 In our treatment, we assumed the classic formulation of the planning problem where actions have no durations. They happen instantaneously, so their effects appear practically at the same times as the actions. Our representation of ordering constraints with CLP(FD) easily enables us to take
18.3 Extend the planning domain definition of Figures 18.1 and 18.2 to a Sokoban-like domain. In this domain, there are boxes in the grid, in addition to robots. Like a robot, a box occupies a whole cell. A robot at a cell adjacent to a box can push the box to the next cell. Unlike with the usual
18.2 Experiment with the POP program of this section in robots-on-grid domains with larger grids and more robots, and observe how the program's execution time depends on the size of the problem.
17.8 Modify the planning state-space definition of Figure 17.7 to introduce the cost of actions: s( Statel, State2, Cost) The cost can, for example, depend on the weight of the object moved and the distance by which it is moved. Use this definition to find minimal cost plans in the blocks world.
17.7 The example heuristic function for the blocks world developed in this chapter simply counts the goals to be achieved. This is very crude as some goals are clearly more difficult than others. For example, it is trivial to achieve on(a,b) if blocks a and b are already clear, whereas it is
17.6 Consider the simple heuristic function for the blocks world of Figure 17.7. Does this function satisfy the condition of the admissibility theorem for best-first search? (See Chapter 12 for the admissibility theorem.)
17.5 Trace the planning process, based on goal regression, for achieving on(a,b) from the initial situation of Figure 17.1. Suppose the plan is: [move(c,a, 2), move(a, 1, b)]
17.4 The natural places where domain-specific planning knowledge can be introduced into our planner are the predicates select and achieves. They select the next goal to be attempted (determining the order in which goals are achieved) and the action to be tried. Redefine these two predicates for the
17.3 Trace by hand the means-ends planning process for achieving on(a, 3) from the initial situation of Figure 17.1.
17.2 Define the planning domain for the cleaning robot of Section 4.2 where the actions are 'go', 'push', 'pickup' and 'drop'.
17.1 Extend the blocks world definition of Figure 17.2 to include other types of objects, such as pyramids, balls and boxes. Add the corresponding extra condition 'safe-to- stack' in the can relation. For example, it is not safe to stack a block on a pyramid; a ball can be safely put into a box,
16.5 Find all evidence sets E that d-separate a and d in Figure 16.3b.
16.4 The program below specifies a Bayesian network according to the conventions used in the program of Figure 16.4. parent(a,c). parent(b,c). parent(b,d). parent(c,e). p(a, 0.1). p(b, 0.2). % p(a) = 0.1 p(c, [a,b], 0.5). p(c, [nota, b], 0.0). p(c, [a, not b], 0.9). p(c, [nota, not b], 0.4). p(d,
16.3 For the Bayesian network of Figure 16.3b, express the conditional probability p(b|c) in terms of probabilities that must be given by the definition of this network.
16.2 Consider the Bayesian network in Figure 16.3b. The structure of the network enables the simplification of some conditional probabilities. For example, p( e | a ^c) = p(e c). Here, the condition part a A c was simplified intoc. In the conditional probabilities below, how can the condition parts
16.1 Often it is possible to make interesting qualitative inferences about probabilities in a Bayesian network without knowing exact numerical values of probabilities in the network. Consider the network in Figure 16.3a. Let both a and c be very rare events, and let all the links in the network
15.4 Let geometric figures be represented as frames. The following clauses represent a square s1 and a rectangle r2 and specify the method for computing the area of a figure: s1(instance_of, square). s1( side, 5). r2( instance of, rectangle). r2(length, 6). r2(width, 4). square( a_kind_of,
15.3 Trace the execution of the query ?-value( ross, relative_size, Value). to make sure you understand clearly how information is passed throughout the frame network in our frame interpreter.
15.2 Let an expert system approximate the probability of the union of two events by the same formula as in our simple uncertainty scheme: P(A or B) = max(p(A),p(B)) Under what condition does this formula give probabilistically correct results? In what situation does the formula make the greatest
15.1 Write a program that combines forward and backward chaining in the style discussed in this section.
14.5 Write a procedure show2( Solution Tree) to display a solution tree found by the andor program of Figure 14.12. Let the display format be similar to that of the show procedure of Figure 14.8, so that show2 can be written as a modification of show, using a different tree representation. Another
14.4 Let an AND/OR graph be defined by: a ---> or:[b/1,c/2]. b> and:[d/1,e/1]. c> and:[e/1,f/1]. d ---> or:[h/6]. e ---> or:[h/2]. f ---> or:[h/4,i/2]. goal(h). goal(i). Draw all the solution trees and calculate their costs. Suppose this graph is searched by the AO* algorithm, where the start node
14.3 Consider some simple two-person, perfect-information game without chance and define its AND/OR representation. Use a depth-first AND/OR search program to find winning strategies in the form of AND/OR trees.
14.2 Define in Prolog an AND/OR space for the Tower of Hanoi problem and use this definition with the search procedures of this section.
14.1 Complete the depth-limited, depth-first AND/OR search program according to the procedure outlined in this section.
13.6 Compare the A*, IDA* and RBFS programs on the eight-puzzle problems. Measure execution times and the number of nodes generated during search, including the regenerated nodes (add node counters to the programs).
13.5 Inheritance of F-values in RBFS saves time (prevents unnecessary regeneration of nodes). Explain how. Hint: Consider the search by RBFS of the binary tree in which for each node its f-value is equal to the depth of the node in the tree. Study the execution of RBFS with and without inheritance
13.3 Consider the route-finding problem of Figure 12.2. How many nodes are generated by A*, IDA* and RBFS on this problem (counting also the regenerated nodes)? 13.4 Consider the state space in Figure 13.5. Let a be the start node and the goal node. Give the order in which nodes are generated
13.2 Apply the IDA" program of Figure 13.1 to the eight puzzle using the definitions of the successor relation s/3 and totdist/3 as in Figure 12.6. Use just total distance (totdist/3) as the heuristic function h (to ensure admissibility). Define also the predicate f/2 (needed by the IDA* program)
13.1 Construct an example state space and the function for which IDA* would not expand nodes in the best-first order.
12.5 Modify the best-first search program of Figure 12.3 to count the number of nodes generated in the search. One easy way is to keep the current number of nodes asserted as a fact, and update it by retract and assert whenever new nodes are generated. Experiment with various heuristic functions
12.4 A mobile robot moves in the x-y plane among obstacles. All the obstacles are rectangles aligned with the x and y axes. The robot can only move in the directions x and y, and is so small that it can be approximated by a point. The robot has to plan collision-free paths between its current
12.3 Let h, h and h3 be three admissible heuristic functions (h;
12.2 The following statement resembles the admissibility theorem: 'For every search problem, if A* finds an optimal solution then h(n) h*(n) for all nodes n in the state space. Is this correct?
12.1 Define the problem-specific relations s, goal and h for the route-finding problem of Figure 12.2. Inspect the behaviour of our A* program on this problem.
11.14 The breadth-first programs of this chapter only check for repeated nodes that appear in the same candidate path. In graphs, a node can be reached by different paths. This is not detected by our programs, which, therefore, duplicate the search below such nodes. Modify the programs of Figures
11.13 Experiment with various search techniques in the blocks world planning problem.
11.12 Study the following search program and describe its search strategy: % search( Start, Path1 - Path2): Find path from start node S to a goal node % Solution path is represented by two lists Path1 and Path2 search(S, P1 P2) :- similar length(P1, P2), goal(G), path2( G, P2, N), % Lists of
11.11 Three search procedures find1, find2 and find3 defined below use different search strategies. Identify these strategies. find1(Node, [Node]) :- goal( Node). find1( Node, [Node | Path]) - s(Node, Nodel), find1(Nodel, Path). find2( Node, Path) :- conc( Path,), find1(Node, Path). find3(Node,
11.10 Sometimes it is beneficial to search bidirectionally; that is, to work from both ends, the start and the goal. The search ends when both ends come together. Define the search space (relations) and the goal relation for a given graph so that our search procedures would, in effect, perform
11.9 How can the search programs of this chapter be used to search in the backward direction; that is, starting from a goal node and progressing toward the start node (or a start node in the case of multiple start nodes)? Hint: redefine the s relation. In what situations would the backward search
11.8 How can the search programs of this section be used for searching from a starting set of nodes instead of a single start node?
11.7 Rewrite the breadth-first program of Figure 11.10 using the difference-pair representation for the list of candidate paths, and show that the result can be the program in Figure 11.11. In Figure 11.11, what is the purpose of the goal: Paths \==Z1 Test what happens if this goal is omitted; use
11.6 Let the state space be a tree with uniform branchingb, and let the solution length bed. For the special case b = 2 and d = 3, how many nodes are generated in the worst case by breadth-first search and by iterative deepening (counting regenerated nodes as well)? Denote by N(b,d) the number of
11.5 Write a procedure show(Situation) to display a problem state, Situation, in the blocks world. Let Situation be a list of stacks, and a stack in turn a list of blocks. The goal show([[a], [e,d], [c,b]]) should display the corresponding situation; for example, as: a e d b ==
11.4 Experiment with the depth-first programs of this section in the blocks world planning problem of Figure 11.1.

Showing 800 - 900 of 4756

First
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Last

Artificial Intelligence In Power System Optimization 1st Edition Weerakorn Ongsakul, Vo Ngoc Dieu - Solutions

Step by Step Answers