Adversarial Search 7 and Games 3.1 + Game Theory, Optimal Decisions in Games, Heuristic Alpha-Beta Tree Search, Monte Carlo Tree Search, Stochastic Games, Partially Observable games, Limitations of Game Search Algorithms playing ‘optimally’ Justify ‘using suitable example. ‘Ana: « Minimax assumes a perfec rational opponent, th so {ates optimal ations However in practice, most human opponent dlpat from rationality. In this cae, the best move a 257 gn Step may not be one that is indicted by Minimax and an igen that takes into consdertion human imperectons vl perform beter. 1 IF the te isto lange to fly explore, then there is posslty that suboptimal move could fake Max into an area of the ft he hadnt considered that is worse than he thought could SPH. but as Stuart said in class, how to deal with suboptimal play it ‘his Stang is lan open research question. Most lgoritns crane chor progae awe an optimal opponent, 22 joes seem to hut them very much-but that's an empl theoretical, result, fb th 7 + Optimality sil well defn, even if your opponent isnt payeE well Moreover the game ti rll enough that your age! fully explore then the optimal player really doesnt care what th her one dows. SS Aaversarial Search and Games Artificial intligence «Lets say Max goes first. What will Max do? He will look at every possible game sequence. He will then take the action which fuarantecs that he will get a score of atleast X. No matter what Min does in subsequent moves, Min can never get a score less than X. So Mav's score will only go up if Min doesnt play well 02 We have two players: MIN who plays first and can make 4 Gases, MAX who playe second and can make 2 moves. Suppose hat efter 1 turm, the values of the leaves are av inthe Fig. Q2.1 crop et sti, nme eo Sa tector sa ari ct dis aia fa ‘Mf we perform an alpha-beta pruning. : ‘Ans, : Now, applying alpha-beta pruning algorithm, Role of Beta (B)- For root A, the minimizer, value ~ 8, is assured by: ‘moving to node D. This value is compared with nodes B, C and E This value of B is now used as reference point. Any node whose vvalue is less than or equal to B value is acceptable else are ‘Teected that is pruned. (As shown in the Fig. Q2.1(a)) Role of alpha (a) - for, root as D, the red >, the maximizer is assured by node having value ~8. Ths is compared with other leaf node Any node "utc or Eaten SenSean, Sener ge. (AS hon in eg 2th une 4 Q Nig mnt 6 A Fig. 02.48) G3 Consider the game tee of Fig. Q.3.1 i scores are from fat players ew. Sup “ om fot lasers point of iw, Sup Dbackedup values in the tre. What ree. What move will the nodes are expanded from left to right, {aan © d4| ) 6 which the ae 08 the fet pax ‘search, shew hy the MAX chow 1 what nodes woud wt ‘be wisted using alpha-beta pruning. “Git or Enger A | Adverarial Seach and Games “ al nteligece ‘Ans. Applying min-max algorithm, ot fax mh dagadadqogoede 7AWHHAAIOAGRHaS of / Applying Alpha-Beta pruning DHAHDHAHBOAHS 4 Explain Minx enrch proce wth an example Ana. 1. Given a game tre the optimal stategy can be detained by examining the minimax value of each node “A Gee for Eninering Suds2 The minimax vale of node isthe wlity (FF player cay Man) of being inthe coresponding state, ASSuming that fog players pay optimally from this stage tothe end ofthe game 4, The minimax vale ofa terminal state is just its uty. 4 Given a choice, MAX will perfer to move to a state of maximum Value, whereas MIN prefers a state of minimum value MiniMax Algorithm : The minimax algorithm computes the minimax decision fom the cen state It 5 used a8 a Searching technique in game problems. The minimax algorithm performs © complete depth rst exploration ofthe game-tree. The Algor 1. The start node is MAX (player 1) node with current board ‘configuration, 2 Expand nodes down (pla) to Some depth of lookahead inthe seme 3. Apply evaluation function at each ofthe lea nodes. 4. "Back up values for “Bask "vale fr ach none nodes un eomputed forthe 5 AUMIN (layer 2) nodes, the backed nx 8p value is the minimum Of the values associated with its children 6 AE MAX node (MAX nodes, the backed up value i the maximum ofthe associated with its children 25 Conder the ftlows whieh following 2 player game sree in Sammast given from the first player's yoint of view ; a roe ch, fist player le the maximiziog player What move oven ? Why ? Use minbmax search to soe mt ‘lain limitations of im paonly nkimax search. HOW 0 overcome, nies liom % 35 nd Games Artificial Intelligence ‘Ans. : A disadvantage of the minimax algorithm is that each board state has to be visited twice : One time to find its children and a second time to evaluate the heuristic value. ‘During the process of searching for the next move, not every move (ie. every:node in the search tree) needs to considered in order to reach a correct decision. In other words, if the move being considered results in a worse outcome than current best ppossible choice, then the first move that the opposition could = “Aut for Engincering SuenArtificial Inteligence ‘Adcersarial Seth and Gang ‘make which is less then the best move will be the last move ‘one need to look at, as the opposition will at least choose thay + Mamas tens to be fo slow for games such a chess. tum the plyer has mary choices to decide on, the brnchay factor of a game of chess is huge and therefore the deeper we ge the slower it ges. On average, the branching factor for chen tends to 30. This is, 30 subtrees per tur are created t } i cr Eng Sa a8 Adversara Search and Games method is better minimx version which ‘minimax algorithm. Alpha beta jed and therefore We “Alpha beta. pruning Gvercomes the Limitations of ojaces the number of nodes to be expands Feduce the branching factor. dae Explain alphacbetn cut off ecarch with an example, State * 28 Ehen to-do alpha pruning: ‘Ans. : cB Pruning a problem with sinimax algorithm search is thatthe name? The Phe mates it as to examine i exponential in the umber of 2. ee propo {eattout looking at every Stops in Alpha-Beta Pruning PeRAX player cuts off search when he knows force @ provably bad outcome 2. MIN player cuts of search when he knows MAX-player can force provably good (for MAX) outcome. 4, Applying an alpha-citoff means we stop search of a particular cern because we see that we already have a better ‘opportunity elsewhere. 4. Applying beta-cutoff means we stop search of = particular Dench because we see that the opponent already has a better ‘opportunity elsewhere. 5. Applying both forms is alpha-beta pruning. a= Pruning Example : Step 1: wes to compute the correct minimax algorithm decision node in the game te. 13 MIN-player can Max move 23 Mun meve 3 7 Re Fig. 2.6.1 Pruning example () |! = Ge for Bnginering Sader,aU Adverearal Search and Gang | gical intelligence 3:10 __Adverearial Search and Games "as 1 Minimax provdure i a game playing strategy in which pam Mopt ose soaegies which wil maximize Wer gas Step 2: wows - Awhile minimizing their losses. 4 The minimax search procedure is a depth-first, depth-limited aN move search procedure, The idea isto start at the current position and 3 32 tice the plausible ~ more generator to generate the set of possible Niecessor positions, Now by applying the static evaluation wax meve oO function to those positions we can choose the best one. After aria xk Going 20 we can back that value upto the starting position to iil Intlgnce C F luation of it. The starting position is exactly as (062 Pruning example represent our eval f os cha Senet food as the postion generated by the best move we can make eae | onext. axon 2a Assuming thatthe static evaluation function retums large value — ihiicate good situation for us, 20 our goal is to maximise the pes ate value of static evaluation function of the next board position. It is 3 ca | simple look bead strategy for two person gameplaying rxmove | a x x ae 8 2 eee anit Fig. 2.8.3 Pruning example (i) woven — iss | teen | the minimax procedure for game playing. Find out the tee —e fT. Filtg of the nodes starting from node A to”. as shown In sing minimax procedure ‘aauming that root node le sunt — [5] = Aa ae A r ee SHON SE a oho m Oa & oo peconee ee SGI catarhin ALIARIC Cai | RAs sge mse wat me vata nin vate wou Fig. 07.1 ——— co Tana eaten | SF 7 apr Brgy Sateiil Intelligence 3-12 Adversrial Search and Games ‘Step 4: Max moves A © fe] td ee Ga gn Gees GECo aa Fig. 7.8 @ Fl ® gy 8 [eh \(H @ (Wl SHS HEE Fig. 73 Final value at root node is value 11. as Explain - f pruning procedure. Mark the nodes in the 28. SB Sasch wt pre out i ang, i EHO SAE i} ae Ana. : Alpha-Beta Pruning, 1 The problem with minimax algorithm search is that the number of game states it has to examine is exponential in the mumber of 2. ai proposes to compute the correct minimax algorithm decision ‘without looking at every node in the game tre. —_—_ 7 “A Gul for Engineering Sten+8 Afversaril ea ng epsmtgee _*® Se a yning wap Pe tM Fae a peo ba utc r search when he knows Max. ape of . a ye (NK ote Ply ee ct ns fp seach of frdh ect we sce tat we already have 4 4! @ opportunity elsewhere ‘apg Bett means we stop search of a tech base we se Bat the opponent already has a hu, opportunity elsewhere 5: Applring both forms is alpha-beta pruning, Solon to pruning for Fig. 81 ole of Alpha (0) - For node A the maximizer, a value of 1 assured by moving to mode B. We c: : _ = an compare this value Cit mit A goin wll palo | spear than 1, Hence, this value 8 being the leat tht ‘maximizing node can obtain, is set a the value of a. ‘ Tis value of is ow use as referenced point. Any node wios vale is sete than is ac ares han a ae plable and all nodes whose vs maa mors, it is known that C node value, Hence entire subtee ely rejected. has value 2 which is tinder node Cis comple afc tligence comparing. The minimizer would be peta ( B) is passed to E for than E. Hence entire tree under E is benefited by only moving to D pruned a9 How can “Ans. : In @ game of chance we e search tree. These no 1m element, ‘uses probability P attached with chance flue. Successor function § (N, di) give minimax also be extended for game of chance ? ‘can add extra level of chance nodes ides have successors which are the ‘outcomes of randor ‘The minimax algorithm rode di based on this ¥ roves from position N for outcome 2.10 What lo game ? ‘ans :4 The term game means a sort of conflict in which n inuividuals oF groups (known as players) participate «+ Game theory denotes games of strategy. acknowledged as father of game theory. ‘John Von Neumann is theory in 1928 and established the Neumann defined game ‘mathermatical framework for all subsequent theoretical developments with other 1 Game theory allows decision-makers (players) to cope decision-maker (players) who have different purposes in mind. In other words, players determine their own strategies in terms of the strategies and goal of their opponent. «+ Games are integral attribute of human beings. Games engage the {intellectual faculties of humans. +f computers are to mimic people they should be able to play games. 11 What is game tree ? ‘Ans. : The game tree : The inital state and the legal moves for teach side, define the game tree for the game. —_——_——— “A Guide for Engineering SudensAdve Search nd Gay Arpt etelignee game tree. A & 7 3 «+ Description ofthe tS Fig. 0111 Example of game tee 1 Root node - Represents board conf ‘hati the bes single ¥ my tum to move, indicating i is my tum, iguraton and decision, required as to next move, then the root is labeled a MAX node 2 Arcs Represent the Possible legal moves for the player that the A Gade for Eninering Sinden Artifical Intelligence 3:16 Adversarial Search and Games 4 Since moves altemate, the nodes at level are of opposite kind from those at level i+ 1 012 What are properties of mintmax ? ‘Ans: Properties of Mini-Max 1. Minimax provides a complete solution for nite tree 2 Minimax provides optimal strategy against an optimal opponent. 3, The time complexity is 0 6" 4k The space complexity is Ofbm) for depth-first exploration where algorithm generates all successor at once, of O(m) for an algorithm that generate sucessors one ata te. 2.13 What te a made-up game ? What 6 ply ? ‘Ans: The ‘made-up! Games and Concept of ‘ly’ ‘For study purpose we can form a game # tree which is constructed up to certain level (Le. upto certain depth). This is called as a made-up game, The term ply refers to depth of the tree. + For example - ply 4is level at depth 4 below the root node. example Ato ply game tee, for game oftctactoe. 1. Assume that two players named MIN and MAX who are playing the game 2. MAX is playing fist. 3. The possible moves for MAX atthe root node are labeled a3 and a3 4. The possible replies toa for MIN are by, by and 50 on. 5. This particular game ends ater one move each by MAX and MIN. 6 In game parlance, we say that this tee is one which, moves deep consisting of tworhalf-moves, each of whichis called a ply A ale for npicig Sensgree 7 N R 3 ROBERT 1. Letinially "Des! so DYD=T" therefore "T=0' & '<5s1. E-9 2 In column 5‘O¥Es0' as ‘TY so E cannot be 0, therefor'ty | p= 7 (0-90 i possible if 21’. Therefore ‘<2=1'. o3 3. tn column 3 'A+A=9, but addition of any 2 same nubei) T = 0 SIS, sen tut adtionis|9 which is poute Wl Q1s sete fatonag crptathmetic problem with appropiate there is carry. Therefore ‘ete’ 0 ‘Af strategy /stepe 4 Remaining numbers to be assigned are (123674: {ONRBLG). 5 We have E47 & 21’ so from column 5 we get ‘ele from columd we have ‘LaLtc5=R’ where ‘Sel therefore I Sh B can be order. As ‘DaG does not genenie ct) A shown in column 6 so R cann ‘oer. 1ot be Jor3. Therefore | ga 6 We have staat ‘Lar'* 7 0 from conn 2 we have ‘LaLete, te APPLE 7. From columa 3 there ‘ng, "BP AYAScte and go there is 0) Here ye know that Tis na 0, a8 ToT i nt equal to T. The case is 8 From column 4 we fame with A. After that we have to solve every letter for the ‘Nera sett NeReetep 27 & 44) problem making each letter unique, The steps ae below. Or therefore ‘Napag ne WE have Ro7 & AE aking a es 10 saisly he wih remain , int “NoBag’ we $$ “A Guide for Engineering Sens Tonite fr Baeas + 9HAS APPL 89 + 99 APPL a9 I + 9HI9 | APPSE 819 +9219 AN ais +919 1038 FINAL | ANSWER: EAT = 819, THAT = 9219, APPLE 10038 e——_ 4 “A Gd for Engineering Swle® Artificial Inteligence 3-20 _Adversaral Search and Games He? p-0 L-3 ‘até Discuss AO* algorithm. Give one example where AO" le friable t0 apply. ‘Ans. © AO? is useful for searching game trees, problem solving Aes put in most cases more domain specific search algorithms (¢. Sipha-beta pruning for game tees, general or domain specific planing algorithms) are used instead. ‘in particular, AI uses Knowledge-intensive approaches and in practical applications heavy use is made of domain specific Knowledge or problem conditions to produce better (faster or more optimal solutions). «+ Game search is an example where full-breadth search is standard, but this may because of the small (relative to other domains) size of the search space. Even in game tree search, extensive use is rade of problem specific features, ie. often search is terminated only in quiescent states (je. not during a forced exchange or when there is a check). ‘In planning, often knowledge is used to guide a search of a ‘generated solution space rather than doing a state space search. ‘This gives non-optimal solutions but for many domains it yields reasonable solutions at much less cost. 117 Solve the following Crypt arithmetic problem : BASE +BALL “A Guide for Engnerng Sodengit sf seesmtene Aversarial Search and Games Adversarial Search and sen_Afeereat a Con a We can say seit _— a distinct digits ee Seis ta M should be either 8 or 9 Fen warasseion! ble cary overs when x= 0 ving 2 nunberss0 possible CAMY OVETS aT ith As we ate just a4 9 when x= 1 Lord. jome carry that is suming M= ne we dng BB ges ws some cay Ca is greg assuming M= 8 Ba an we et ait mmbrs odd uP FM 8 BE da ay x 0 sanber seLty-E Hence, G cat be 280 eat eye 506-1 SeLey+Lelotyss ike this Now ou question wil ok somewhat ik Pe BASE | where y can take either 0 of 1 lane | ify = 0 then L = 0 then S$ and E are not distint which is not a | possible according to question. 1 AMES p tye The then L = 45 which is again not possible as L must be a single digit if 2B are added and they are giving us value greater than equal] TOE 9 to 9 So, M=8 is not possible 10 Then, B must be greater than equal to 5 Sojour question further reduces to 1 Consider B = 9 ist 74SE + 74LL 4B =9 then A= 8 which snot possible as then B + B + 1is rl equal to A ifB 8 then A =6 which is not possible for same reason Take B ~7 then A» 4 which is possible which is as itis not leaving 287 ‘carry over. Gan 50, 14988 Now, S+L=101+E=10+E, and E+Letty+s So, hhere y can take maximum and minimum values of 1 and 0 respectively, ——$—_——______. <= “A Guide fr Engineering Stier carry over from E + L ‘rowing the fat that x and y can take maximum value of 1 204 ‘minimum value of 0 “A Gute for Engineering Soeia Advesrial Search and Can pl tee Ey ayo Then, E+Les or Sob* and S+L=10+E Q Adding both we Be | rg=2E+10 E oy SeE+5 Subang equation from 2 we get mL=10 oLes Soour question farther reduces to 10 14S +785 HSS %, E45=5,and S+5+0=E +01 S+5=E+10 S=Es5 Now we have to choose athe than 1479 and SES HOMO, 1, 2 3, 4,5, 6, 7, 8 and) Some cn take va lke vats of E and S fom 023,68 such that § = B*8 Os there 5 cant be chosen 2s there 7 cant be chosen “A Ge for Enginerng Sui Artifical Intlignce Adversaral Search and Games 6 is there 11 can't be chosen 8 is there 13 cant be chosen so, only possible pair left is 3 and 8 which satisfy our constraints. Hence, $=8and E=3 10 | and Meo 18 Solve the crypt - arithmetic problem with the followin Constraints. Give solution atepe, Sonatas 1) Use decimal arthmetl t) No two lates posese “A Gale for Engineering SuenAdverse Search ond Gang 18 angsariont an Hee OS 62513 a ae o-n we “5, Cra 30)+5}=RI6) | ' waoss $0) Di1}=€14) sna he 8 02) +418)= 67) Ie) +42) = a) uatces C18} +RI6)= 1+ Cary 5) | eta Artifical Intligence 19 What do you mean by constraint satisfaction problems ? Explain constraint propagation algorithm using suitable example ‘Ans. : 1. A constraint satisfaction problem is defined by a set of Variables Xy,Xp,.,Xq and a set of constraints Cy,Cy,..., Cy. 2. Each variable X, has a nonempty domain D, of possible values, 53. Each constraint C, involves some subset of the variables and specifies the allowable combinations of values for that subset. 4A slate ofthe problem is defined by an assignment of values to some or all of the variables. (X; = Vi,X) = Vj.) 5. An assignment that does not violate any constrains is called a consistent or legal assignment. 6A complete assignment is one in which every variable is mantained and a solution to a CSP is, a complete assignment that satisfies all the constraints 7. Some CSPs also require a solution that maximizes an objective function Consider graph colouring problem as shown in Fig. Q.192. Constraints are - © Fig. 0.19.1 Graph colouring problem 1. We have three colours for colouring a vertex. 2. No two adjacent vertices have same colour. —_— “A Gude for Engineering Soe a5 “4 Golde forEnpnering SernAdversarial Search and Gay, one ‘Allowable combination fo, approsh popes information from assigned ty 2. The forward CNG Po avoid or detect all failure. ales ba anton peel efores constants lea, a0 : fast method of J he sof arc conssterey proves 3 of constraint propagation ha ssa STONES than forward checking re festa deed arin the constraint graph | constant Propagation using Are Consistency | 11 This fast method of constraint propagation. | 2 X 4+ Y is consistent if (for every value of X there is some | ‘owed vale Y. For example (VV, i8 consistent if Vy +Re, Vy = Bu) efor evry value of x in X there is some | lowed valve y in Y}. This is directed property example - | Mey | Vy = Vp is consistent iff | V, = Red and V; = Blue as mcd ares between variables represent the domains of 4 Conn ‘Yarables, they are consistent with each other. raint propagation i Fegan cm he ppd as prepraceing | pial Intelligence 3-8 ‘constraint propagation using 1. Arc consistency is not caps Partial assignments [V = Red, V2 2. Keconsistency is very strong form 3, A CSP is K-consistency any consistent assignment to those variable a ‘Adversarial Search and Games K-consistency able of detecing all inconsistencies. = Red] are inconsistent. of constraint propagation. if for any set of K-1 variables and for consistent value ‘can always be assigned to any K" variable, END... ‘A Guide for Engineering Sale <= a “A Gale for Enincering StatenMagn 2 Snevin a2 Knowledge rifical nelgence Pompeian (x) indicates that x is @ pompeian. Next symbol "+ is ‘implies ie, means and roman (x) indicates that is Roman In other words we say that all x i. all pompeians were Roman, iy Caesar was ruler ruler (Coesar) Here, one can ignore the fact that proper names are often not references to unique individuals, since many people share the same name. Sometimes desiding which of several people of the same name is being refered to in a particular statement may require «fair mount of knowledge and reasoning iy “All romans were ether loyal to Caesar or hated him ‘x: Roman (x) = loyalto (x, Caesar) vate (x, Caesat) ln English, the word "OR" sometimes metns the logical inclsive-OR and sometimes means the logical exclusive-OR (XOR), Here, inclusive interpretation is used. Bu, if one considers it as ‘exclusive-OR’ then presentation would be Some what different a follows = x Roman (x) ~» [loyal (x, Caesat) v hate (x, Caesar) 0 4m complex sal word problems sich as shopping on internet orbit ~Uoyalto (x, Cats)» hate (x, Caesay} se_emeene. pe ‘nore “gral an. 2th) scpene sl oa eesti, We ned 10 eee ropesenition® which concentate a general concepts such as Actions, Time, Physical Objects and Bel 5y | loyato (9). Sch nos many ince Psi Mere, can we Both the quis Le. + > univer auntie and 3 existential quanti. 2 Represent positional ogc. Present following facts using propositional logic. So, one can read above presentation as for all x, there exists a y such emis) Ca a al that loyal toy of indietly we say wer ther loyal t Cassar or bated him. e : in) Everyone loyal to some a. "Everyone is loyal to someone ©) People ely 1 ana rar i But here one can understand a diferent meaning i, ther is someone oae rulers they are no loyal. i) Mares tried ta ‘0 whom everyone is loyal, whi vwiten as P Marcos tried snasinate cease. earprus oa, ante dl © WHOM everyone is loyal, which would be wien 3y : vx : loyal (x, y) Am. 1) Al pompeins were Romans “*: Pompeian (2) + Roman (a) "Now read it as, there exists y, such that forall x, x is loyal toy. Or tn tis presentation iver "aie OF 1, forall sm aiege * # Wed 0 presen ints, Knowledge-Based Agents, The ce hepa Ts | roving Efecive Propositional Model | bed en een Logie | {24 How knowledge cam be representation ? Explain with example, EE (SPPU + bert Ans: Kaowiedge Representation ~ In toy problems the choice representation isnot imperant because these problems have con vcabuly “everyone is loyal to y" 1) "People only try to assassinate rulers they are not loyal to coors A Gude for Engineering Sudentsee ‘Syntax : 1 defines allowable sentences inthe model, Propositional esleulus symbols : Se Na es ed ge 4. Symbota ee eer mere eee FORKS on tenets srs 4) Predicate symbols: They ae used to represent relition “efor Enger SnFor example * wa se we wil ee sent 8 Ive te READS IPH BO ym i EEO TEESE oj nis may be Physica anes ne Mc ees mS ee ee oT eo pa coer frm Dip a BOOK ae eau For eae: sya cvs ike FA 0 SYBOL, Thy ay iy te mt bokeh tty 1 10 be rei | ses be el Sep READ (>) in the domain of disco Tey denote incions +) Fan sys: Ted ae ote is mai ie for example: "Ras fore woud be, MARRIED (MOTHER (Ram), FATHER (Ram) where MOTHER, FATHER and MARRIED are ftion symbols 2. Truth symbols Tine, Fae 3. Comectves | Not: Pepston! smtals dnote prepositions, or statements abou word tt may be ether tue or fuse, such a5 "the car is red" or "wit is ver. 15 What are propositional calculus sentences Ans. ') Every prepostona smal an truth symbol is sentence. For example : Tr, P,Q, R are sentences, 5) omic semen It i indivisible (aon-composite) sit) slement, T consis of preposition symbol, Each such 5 stands for a Preposton tat can be tus or false ') Complex Seine 1s cesrated from simpler sents logical connectives, Artifical Inteligance 4s Knowledge ae rodge Note : Logical sentences are also called as well-formed formulas or WEF, 26 Explain inference algorithm in propositional logic. Ans AA logical inferencing means to decide whether KB kc. for some sentence General Inferoncing algorithm (truth table enumeration algorithm) 1) This algorithm is used for deciding entailment in prepositional log. 2) It works like backtracking search algorithm, 3) It performs a recursive enumeration of a finite space of assignments to ‘tables, 4 This algorithm is sound, because it implements directly the definition of entailment, This algorithm is complete, because it works for any knowledge base and ‘ct, and always terminates (As there are finite models to check). Time complexity of algorithm is O(2") (assuming n symbol in all, then there are 2 models) 1) Space complexity of algorithm is O(a), depth first. 6 because the enumeration is Note: Every known inference algorithm for prepositional logic has a worst ease time complexity which is exponential inthe sizeof the inp Note : Prepositonal entailment is CO-NP-complete. (CO-NP stands for Complement ~ Nonpolynomial time, where as CO-NP-complete means hardest problem in CO-NP) ‘The functions for truth table enumeration algorithm : Function TT Entalls (KB, o) Retums True or False Inputs: (KB, the knowledge base), (@ sentence in prepositional logic a), the query, a sentence in prepositional logic. Symbols A ist of the preposition symbols in KB and & ‘eum TT CHECK-ALL (KB, o, symbols [] ) Recess “A Gale or Egiering Sendenx deciding preposition 4) PLU west: eens is iin # mel 2 wie el eens 2 pail mdelan assignment 0 eh some ofthe arable | 4) The foo eal EXTEND (P, tue, model) retums a new par collin wich P has the valet 7 Represeat Wampus word wing propositional logic ale inference proces. (OR Write the algorithm for Wampus world agent to find pis wampses and safe square, oR Expai ia Propositional loge, fa: A wumpas nord agent: ape Word iso 7 pus oe ‘ficial intelligence problem, which is = oo ‘bused on simulation, as well as oe Wamp nat an ety comp sv ade op fo Se ace based agent in Wumpus world based % par fe in which an agent had to exBlie interconnected rooms. In one ofthe 55 wot “A Gd for Enginering SH tfc Intelligence a8 Knowledge cate cave there was a Wumpus which would kil the gent if tener ia om. Some rooms contained pis, and the agent would die ff tht td any of those rooms too. Tbe agent had one arrow with which i ‘The goal was to locate the gold that was hidden ould Kill the Wumpus. r sete wherein the cave and return to the start without geting Killed «soe jee 2 Sech is, A Fig. Q.7.1 A wumpus world agent and its environment PEAS description : + Performance measure : Gold : + 1000, Death : - 1000 ~ I per step, —10 for using the arrow. + Environment : + Squares adjacent to Wumpus are smelly + Squares adjacent to pit are breezy. Glitter iff gold is in the same square. Shooting ills Wumpus if you ae facing it, It screams, a = Shooting ses only arrow. ~ Grabbing picks up god if in same square Releasing drops the gol in same square = You bump if you walk into wall. — =) “Gale for Eninecrng Sader2) sae ys-W 3) Dist ws 4) Sage ves Wasi esily 0 natural fate ing the wurpus word | Ess ge neal consis te res ofthe envionment 2 in pe Oe, Bap, Se] Se a aces aS ee G = Gilter gold OK ‘Stench a fat an Tat Visited ‘Wumpus Laxlo{ | | Fig. Q72 Exploring wumpus world-1 Action : Move to safe cel (2,1) 3) Location: 2,1} _2eo > Senc,Brez,~ Gite, Bump, — Sereat) eo Atal Intelligence “0 Knowledge Artificial mtelligenee Ar infer : Breeze indicates that there is a pit in (2, 2] or (3,1) ion Return to (I, 1) ty next safe el ae pe eT 3 = Greco | ps pe ps fas G = Glitter gold ok = Sof equa aa ae eee = ime Bs 3 = Stonch | omg ea Vo = Viiaa v |e w= Wonpus | | ox | Fig. @.7.3 Exploring wumpus world - 11 4) Location : (1, 2] Percept: (Stench, Breeze, — Gliter, + Bump, — Scream], Inger : Wampus in (1, 3} oF [2,2] YET... notin (1, 1] “Thus. notin (2, 2} oF stench would have been detected in [2, 1) Thus .. Wampus is in (1,3) [2,2] is safe because of lack of breeze in(,2] Thus pit in (3, 1] Action : Move to next safe cell [2,2 ia Bae ae D = Aen is fess as pee wh & = Gitergou Ok = Sot square fag 22 [aa P= Pt S ox | o« oe lon it te aa [at Y= Vites wp ape Wo = Wompus St a Fig. Q.7.4 Exploring wumpus world - I cess A Gul for Engineering SatetsWorld “4 urs like a general ~ Agent sled bse HAE & BNET oy inorance based ABO od "el inet kes tase aE 7 sword ay about te Fain of Bits (P), a sl 1) Tas an (> oa sale ates the "physige ma se oe Oe Physics 2 bis ” snp wth seaon fom a squate KNOWN t0 be safe, 4) Tega bis 187 =Ry:3 “aio 4 Fereey st eee ; Sr Beg Payer Pet Party Y Peay Seen apace yet sea -Ryg Ruy © Buy 6 Payot ¥ Paget ¥ Party ¥ Panty) 6) How dow ees ee is erty one Wampas ? Wea ese in 2 eMeE, 2) Tae at et ne Was Rs Way Wave Waa Fara mo sures ne mus be Wamp, i) Ts est fr ey pif squaes ii o(o-1))2 numberof sentences Rye Wy Wy Ry WV Whs “Ris 7 We hve wel tt aol inthe mowledige bate urpus-Werégent program 1) ASK-Easinest compan TT-Entils would have to enumerate 24 DPLLWALKSAT petonm A Gude for Engineering et fia Inteligence “2 Knowledge ) For a large Wumpus world, the number of sentences in the knowledge base will be huge ) We could not capture the propery “breeze in a square indicates pit in atleast one of directly adjacent square” ation, entation In wumpus world Of ty 1) Can we add proposition like, Lj, * Facing Right » Forward > La. 2) Knowledge base will ental both Ly, and Lz, by using inference nes 5) What we intended to captre is this FU 9 Facing Right « Forward +L Facing Right « Tum Left ) — Facing Up (+) 14) Fox every times sep, one such statement 22? Assuming the problem is solvable in 100 times steps, then 10 thousands of addtional sentences are needed 8 Explain how to have inference using propositional logic. Also plain resolution uslg propositions! logi. s.£ 1) Modus ponens is represented as, = Bia B ‘means that whenever any sentences ofthe form a= lsd care given then the sentence canbe infeed on) [For example - ‘Wo statements, (Signal Pole Ahead » Signal Red) = Stop ) (Signal Pole Abead » Sigal Red) are given then "Stop" can be And-elimination ; It says that ffom conjunction any conjunets ean be Ted. is represented as, “A Guide for Eninering Suenog ” eo Knowledge exo po yA sentence epee conjunction of junction of lita isd i cave a to be in Conjunctive Normal Form (CNF). rt se isu fr ening ste oot eps for converting preposional logic seatences into CNF ieee Fim, epiing em i (= B-2) eee eae eaeee ees eas cate amn VO it By 2 Pi2v Pai) a jin of ne uta stcion SMS gi he ec, os (By, 1 9 Py,2¥ Paya (Py,2¥ Pa.) Baa) Eliminate —, replacing 8 with - av B (By av Ph.2¥ Poa) Alv(Py,2)¥ Pay Biv) CONF requires —to appear only in literals, so we "move ~ iawards" by hy. aes vier hed seta od mare complementary Titeal. (1 mea 1 oiteers lowin equivalences :~ pSeepsst ate can te genni gy “Tanai fhe lowing ded equal fp oven: The wt sion le -\( 6) = 4 (clouble ~ negation elimination) reso (48) = Gay oP) (De Morgan) yy amy oo Mn avy | Ty isc aa =P De Moran V1 MeV oA Mig In the example, we requir just one aplication ofthe last rue. (Biav Pav Pad a U2 Pe vy By) ‘Now we have a sentence containing nested © and v operators applied lids | to literals. We apply the distibutvity law, distributing v over Serra | wherever possible. : (6 Buy Ph.2v Pai) AGPL.2 (Pai Biv By,1) ‘That is, resotion takes two clauses and produces a new » containing all the literals of the two original clauses except *" Arove semtence is now in CNF. ee Teorey WY he adm ae complementary ites. IF we are dealing cay Clases of length two we could write this 28 comple tes mh olution rule forms the basis for a family of complete inference Ii a single iene rie, tat yields a complete inference dures. when coupled wih ay complet seach algorithm. Clablte da ate ‘Refutation completeness: Gerace LI gaia ca tele ta cal al CcTirn iin ee in ont ie mae eolaton agri ery Sentence of sD) takes, postin lgie is vale input as knowledge base [a sentence in prepositional logic] Conjunction of dsuncons af tents PAY 8 and cc (it is query - which is a sentence in prepositional logic). A Guide for Enginerng Sudens16 Knowledge 4h ridge base do a tiger ‘pt taowtedae Se do ayy, SY ari ee It has atmest one positive literal. Hor clause with exactly one position literal are called definite litera ‘The positive literal is called HEAD of the clause The negative itera is called as BODY of the clause ‘A definite clause with no negative iterals is called as fact. * Hom clause with no postive literals can be writen as an implication whose conclusion is the literal false, Such hom clause can be used to represent integrity constants (conditions) in the So or averted in CNF pd a api it eg CES ins complenenaty TIS aay ined me ito set iit ist aleady prese, isa nse wy Ti anil one of the two 4) Sip (i) 0 G8) ae Pete databases, ses 0 in whic 5) lnfrensing with HORN clause - For inferecing with Hom clase enw ins tha cn be eed 0 WICH ae gg Er Ee tes : Forward Chaining Method eu on resotion rule derives the empty ty An apliain of he rst ta ‘The concept : 1) Iris general concept of data-driven reasoning in which the focus of attention stars with known data, 2) It can be used within an agent to derive conclusions fom incoming percepts. 3) I is used for determining whether a single proposition symbol Q (the query) is entailed by the knowledge base of Hom clauses, 4) It initially stars fiom known facts (positive literals) in the knowledge base, 5) Ital the premises of an implication are known then its conclusion is added to the set of known facts, For example : TE Ly and breeze are known and (Lj 0 Breeze) => By, isin the KB then By, can be added, ©) This process continues until the query Q is added or until no further inferences can be made, Forward chaining algorithm Function PL-FC-ENTAILS ? (KB, q) returns true or fal Inputs : KB, the Imowledge base, a set of propositional Horn clauses, the quory, a — ‘4 Gude for Enincering Suen whic ase celal B Completeness of resolution The algorithm of resotion i complete. It means that it wi poacs a result checking all clauses in KB and all the clauses, ffom them, It finds closue set of clauses set which has all clauses fom exiing clauses set It checks the input query aging closure set Ground resolution theorem 4 states tht ~ "fa set of clauses is unsatisfiable, then the lou of tine clawes cnn the empty clause. The ground reson trem is called as completeness thee rection Boia forward chaining concept and algorithm aloag As: Tho Hom Clause Uta fled Contains simple restricted clauses called coe epee” SH i «disjunction of literal of which #Know ler ol ech Hor cous coun ig HEAD fel = ust (EAD el, agen) sen rps fran caning gr 1) Forward caising algorithm 38 sound ren tt very nln is eset eos 2) Frat hig expe 1 meas at cey eatiled sonic sentence will be dein Algorithm constructs infered table for final state. In this ‘able cath fl el ig poss eval and fae fi al 5) Fr ewope Cowie aa ple knowledge base of Hom clause, as given below. 2BACoA Gally an application of modu DiBsc el nels 418 Knowledge “AND - OR graph for above Horn cause knowledge base 1) Mulple Kinks joined by are indiate = conjunction 2) Multiple links without an are indicate 1 disjunction. 3) The links should be proved. 14) Inforce propagates up the graph 4) Before proceeding ahead, from conjunction point, the propagation carte until all the conjuncts are Fig, 91 AND-OR graph for Hom clause KB known, The order of clauses considered by forward chs inferencing query Q: 1) E and D are given tre. 2) E and D implies B also B and A implies B. 3) D and B implies C. 4) B and C implies A. 5) A implies Q. 6) Therefore Q is tue. a.10 Explain backward chaining concept and algorithm. ‘Ans. +The concept : 1) Tt works backwards from the query i.e if query Q is known to be tut then no work is needed. 2) Algorithm finds those implications in the conclude Q. If all the premises of one of those implications can be proved tue (by ‘backward chaining) then Q is true. 1) Backward chaining works down the graph starting at Q until it reaches 1 set of known facts that forms the basis for « proof 4) Backward chaining is a form of goal-directed reasoning. 5) Iris useful for answering questions such as "what action should T do sow"? — —————— Sco 14 Gude for Engineering Stents ining algorithm for Knowledge base thatwe iil ee eS = ard iin es only relevant facts), "2 : 6c wat ees eso smple : : sesame mowed BAC considagy | in he thee clawes (A v ~B), (By 0), (Cv A) A and B ae pure, Cis impure. sao ree Make a pure symbol literal tue ein) chaining Sed scone by baka Algo 1 eee ey a tsk SO) = Unit clause heuristic : ee Unit clause: Tes a clase which have only one feral in the clause. isi oa ee ‘The ony literal ina uit clause must be te. aera ty Bad Aas well a E and D. + If the model contains B = false, then (B v -C) becomes 3 3 isp by clase becaute iti equivalent 10 (false v —C) or just sC. 4. Cis by Eat ‘The DPLL algorithm for checking satisibilty of a sentence In Sand D agente in knoledge base. pe een 6 Tere fo Q we od iplaons Sequence that conclu g | Function DPLL-SATISFIABLE (6) returns trv o false Inputs: S, «sentence in prepositional logic Clause ~The aot of clauses in the CNF representation ofS. Therefore Q is ve {211 Discus backtracking algorithm used in propositional logic, (OR Discuss Davis Putaum Lopmann-Loveland Algorithm (DPLL). symbols «A list ofthe preposition symbols in S. retum DPLL (clauses, symbole, ||) Function DPLL, (clauses, symbols, model) returns true or false ‘fevery clause in clauses ie true in model then return true if some clause in clauses is false ‘Ans: The concept: 1 Takes apa a sentence in conjunctive normal form, which isa of ss 2) Ie douse dept fst emerson of possible models 3H dtemines if ipa propositional logic sentence (in CNF) i sata, PLLA improved a ‘in model then return false — P, value « FIND-PURE-SYMBOL (symbol ‘model) This ler fir pop ante eater general inferencing algo inpovements oe eas ~ Entals algorithm). There are three desrbed been, Swameraton (TT ~ entails algorithm) 8 1 rnin 1 Ae ay ng scm eye 2) Pure symbo} ‘heuristic : we ‘iP is non-null then retum DPLL (clauses, symbols, -P, EXTEND (P, value, model) P value «- FIND-UNIT-CLAUSE (clauses, model) \fP is non-null then retum DPLL (clauses, symbols-P, EXTEND (D, value, model) P FIRST (eymbols); ost REST (symbols) return DPLL (clauses, r0st, EXTEND (P,true, model)) or DPLL (clauses, res, * Pure symbot a "W905 appears with the same "Si EXTEND (, false, mode!)a ying gg fiction which ef sh Teta 0 1 To “sues FN i 2s fetion which 0 ga USE rng f ) Outputs are registers corresponding to actions }) Circuits are evaluated ina data low model LUNI sll iy FN nate) 5) Value stored at each preposition symbol gives the truth value of the ase OT ppt TOF CHECKING satin, |) comspondingsymbl atte cen tine ae Oe es et ee ties rece al ‘state given a stream of percepts. Walesa 2 aor Jementation for it }) For logic based systems : Maintain a representation of the set of all a atin sen ‘asta, logically possible world states, given axioms and percepts. nis: yaLKSAT algorithm = ) Basic trick + Suecessor-state axioms define truth of preposition at +1 carats ‘from prepositions at t aise oe For example 2 en p's irsonlict Hes of mining Alive! e» Seream! » Alive! ees a aeeeetamt aan os emeen ectness and randOrIDesS ena aA : _ el when we expect 2 solution to exist. for eg 9 (Ly «(Facing Down" 4 Forward*“!)) 4) his ee : it a Yolen (U5 acing Left“! Forward!) 5) Its incomplete alge. Ik return satisfying model when it, the nfea i fit ha poblem, When algorithm fails to ah reve i ay ent int infinite Toop because of infinite vt acl 6) exh seach alg ke WALKSAT, cannot aay zai, wich is very necessary for deciding entaime, Fer example ‘As age cat rely ws local serch to prove that a sau i in he Was wer 13 Exphia circuit based get in Wumpus world bas! ropositional lop A: Ths ae y fiance M® MSA chat ogi expession diet Beeze O genet O iter Fig. 0.134 Part of a circult based agent forthe wumpus word. Figure depts circuit for grabbing the gold andthe circuit for ‘determing whether the wumpus is ave Charts of chet bed ; Hells agent wih ste, sown eg it meena ‘ots are inputs to sequential cireyi. delays are shown as small. has network of gates and registers, “ie for Erineig Skene “ute for Engineen e Knowl “ere again Oeil act) and has 00 ‘moved, tried but bump, 4 en and moved forward, same tv ' Bump") i «The agent was i was at [1,2 sing eas at [2 1 fing Ft Uh <> Uys Forward (ted a (Facing Down!” 1 9 Forward!) (us! Facing Let!” 9 Forward!" ') a Fowaed wt (enter) ao OD Tar] am — sem aaa} “Peon A Fig. 0.122 The cult for determining ifthe agent Is at [1,1]. Exch locaton and orentation register has a similar circult attact A pot oe to heute agent ‘itil ent ale coin in, ‘The agent cannot se a tth value for it : tific Intelligence au 1) Conciseness Krowletge Aas. IBA ~ It require seperate copies of its knowledge base for every time sep ‘CBA ~ It do Hot require seperate copies ofits knowledge base for every tame step. ih agents requires phys sure and 5 engeed a sexes or Seis) fo rey thf ey went se fr le see oanen jy computatona tlney Bee aie erpeunlal tne einer I cil eset omanber of yl) (CBA - It takes linear time for inferencing which is dependent on circuit 3) Completonoss Wa: (Time grows, 1) Iis complete but with huge time requirements for completeness. 2) I also stores previous states/percept therefore memory requirement is also more. cpa 1) Its incomplete for various reasons 8) Acyclicty restriction in circuit itself which is practically not possibles, 4b) For a complete circuit CBA will take exponential time with respect to circuit size, 2) It forgets all previous states therefore can not draw conclusions based (on previous states/percepts. 4) Ease of construction IBA - As environment and conditions are simple and clear. It is easy to Tepresent in propositional logic language. CBA - The envionment considered is limited therefore a circuit is small acyclic and semicomplete. Hence describing it is easy when relation Detween percepts and action is simple and straightforward, It is easy to describe and construct knowledgebase for such agents aac ee RESCH EEE ‘Gacooay Geo ee; ledge artificial Intelligence 4-26 Knowledge 3, It can take up value from allowable objects, relation or function set. || A term without variable is called as ground term. ji) The sentence ¥xP , Say that P is true for every object x, where P is a logical expression. For example : a) “All Princess are Person" can be represented as, x Princess (x) => Person (x) b) "If the universe of discourse is people, then this means that everyone is happy" can be represented as, v x Happy (x) iv) '¥ can make statement about every object. y) '=" is natural connective to use with '¥'. vi) Common mistake to avoid : Do not use ~ as the main connective with V. For example : v x At (x, IT) ~ Smart (x) It means that "Everyone is at IIT and everyone is smart.” It is mistake because you wanted to say "everyone at IIT is smart.” Q.17 What is existential quantifier ? Ans. : Existential Quantifier (3 ) i) These are used to make statement about some objects and not just for all, ii) The objects are not named, iii) The sentence, 3 x P says that P is true for atleast one object x. For example : ‘) "Princess Cindrella has crown on her head" can be expressed as, 3x Crown (x) a OnHead (x, Cindrella) [Crown is a crown and it is on Cindrella's head] b) "If the universe of discourse is People, then this means there is at least one happy person", can be expressed as, 3x Happy () @uconrepe a pat _—_— x such that...” wtoere exists itis pronounced oR : por same Xo" + is natural conned ‘i Common mistake 1 wih 3. wis Fox empl 3x At(e UD = Smart a ‘48 Teast thse sete 1 Je ike al ind of 008 2 Apple ae fod 23. Chicken is food. “Anything anyone ents and isn't Kiled-by is food. Bil eas peanuts distil ave. 6 Sees eveything BU ats ‘Aas: Tete execs int oul in predate logic 1 et ke lin fd "Y Liks (on, x) Food (x) OR / x food (x) — like (John, x) 2) Apple are food. Fou Ape 3) Cie sot fod (itn 4) Antigen as di ed by fod TREE) A= food Q) ) i tue if there is anyone Who is no, ces into formulas in predicate logic. mh a ae iy ) Se everyting Bil ents “8 (Se) eat (Bil, x) anil Intelignce 48 Knowledge 0.19 Convert the formulas derived in above question 18 into clauses. Prove that John likes peanuts using resolution. ‘Ans. : Conversion of formulas derived in Q.19.1 into clauses. ike Uahn, Peansts)] [= Food (Peanuts) v ike John, Pana) | Food (Peanus) False Contradiction Fig. 2.194 1) Ye food (x) like (John, x) ~ food (x) like (John, x) i) 2) Im above equation (i) we substitute x = Peanuts ~ food (Peanuts) v like (John, Peanuts) O) We need to prove, “John likes Peanuts’ ie, like (john, Peanuts) We prove by resolution, we add negation of goal / result in out set of statements, ~ like John, Peanuts) ii) Resolving equations (ii) and (i) Hence, goal ~ like (John, Peanuts) is false. Hence, the proof like (John, Peanuts) 0.20 Using the formulas derived in Q.18 prove John likes peanuts ‘using forward chaining as well as backward chaining, ‘Ans. : Using wif derived in QS. We prove now that, = "Tohn like Peanuts! using forward chaining = "ohn like Peanuts! using backward chainingSova Fil teligence 90 Krowiede “ans. : Consider argument, = Socrates is 8 man this can be represented as, ‘MAN (Socrates) = Men are widely distributed ovgr the earth is represented a5, oa Poo) DISTRIBUTED - ON - EARTH - MAN Ee ror Sets wily dae ne as don, Pou earth is represented as, | DISTRIBUTED - ON - EARTH (Socrates) re (Hence the proof In above case it fils to capture the relationship between any individual ‘az04 ting ‘ta! and that individual being diseibuted on earth, To do tat there ne is need of variable and quantification unless there is need to write vas tn, Peas) Seperate statements about the ‘distibutonship’ of ‘every man’ known present on earth -=+ The facts described by these sentences can be represented a5 a set of wif in predicate logic as Foaa (Pent) ‘sm eS 0 Sn i ert be fats Paanus) Kile (y, Peanuts) mi 0) ed oer) i) Soerates is @ man MAN (Socrates) ii) Socrates is widely distributed over the earth. xs (8, Peanuts) —_—Kied (8, Peanuts) MAN (Socrates) -» distributed - over - earth | | (Socrates). Ae ue 2.22 Explain the algritim of preicate logic resslation tree pean Ant: The resoltion inference rule Fig. 0.202 1) Two clauses, which are assumed are to be standardized apart so that 221 What is wrong wit they share no variables, can be resolved if they contain complementary Sores tape they oc tied oe the earth 2) Firat order literals ae complementary if one unifies wih the negation mn the oer. Their, Socrates : Seca wi 3 ference rule is eae Bee represen : eVincW tae “stn le vo that thin pratt gptiese sentences YF SUBST OI WoW V, has VoWh Ven. VosVng Wager)on al te ee a ete Oe ye Loves (9) sill I pecoapennny a Les (ES) 3) tod Loy ace te resalvent clase ‘ils (0 en is the Bary resolution rule, becage ‘he binary resolution re by ise el ‘5,1: Inference in First-Order Logie, Propositional vs. First-Order Inference, Unification and First-Order naj ane Pae Bee eer foaagi eee acme (ce ett ot gf [nen ns ne co ite . : rer ppoxh is end actrig. Factoring Means the remon 9 dae ern inode cse.Propostional factoring red ‘Re teal o one if hey are identical; frst order factoring reduc oe oe i heya nial Soe prvi by eatin ein pes tt KB F by proving KB» — a unsatisiable, ig Oy dag apy clase Ha prof By contraction Fai a ken neon ot, 1) Coen al probe statements to fst order logi. 2 Coe ot ene sein conuntvenonoal form. 2) Asset te oegton of the goa (1 Consider the following fact 1) The member of the St. Bridge club are Joe, Sally, Bill and Ellen. i Joe is married to Sally. ii) Bill & Ellen's brother. Jn) The spouse of every married person in the club is also in the 7b ) The last meeting ofthe club was at Joe's house, 1) Translate the above sentences into formulas in predicate logic. 2) Prove that Ellen is not married, ‘Aas. : (step 1) Predicate logic 1) MEMBER(oe, ElmStBridgeClub) [MEMBER(Sally, FlnStBridgeClub) MEMBER(Ellen, ElmStBridgeChub) MEMBERGill, EImStBridgeClub) 2) MARRIED(oe, Sally) 3) BROTHER(Ellen, Bill) 4) Vix vy MARRIED( MEMBER , ElmstBridgeChub)) 5) LastMeetingtHouse(Joe))