Artificial Intelligence 06.2 More on Causality Bayesian Networks

1. Artiﬁcial Intelligence Causality in Bayesian Networks Andres Mendez-Vazquez March 14, 2016 1 / 112

2. Outline 1 Causality Example Definition of Causal Structure Causal Networks Causal Chains Common Causes Common Effect 2 Analyze the Graph Going Further D-Separation Paths Blocking Definition of D-Separation 3 Algorithms to Find D-Separations Introduction Example of Reachability Reachability Algorithm Analysis D-Separation Finding Algorithm Analysis Example of D-Separation Application 4 Final Remarks Encoding Causality 2 / 112

4. Causality What do we naturally? A way of structuring a situation for reasoning under uncertainty is to construct a graph representing causal relations between events. Example of events with possible outputs Fuel? {Yes, No} Clean Spark Plugs? {full,1/2, empty} Start? {Yes, No} 4 / 112

5. Causality What do we naturally? A way of structuring a situation for reasoning under uncertainty is to construct a graph representing causal relations between events. Example of events with possible outputs Fuel? {Yes, No} Clean Spark Plugs? {full,1/2, empty} Start? {Yes, No} 4 / 112

6. Causality We know We know that the state of Fuel? and the state of Clean Spark Plugs? have a causal impact on the state of Start?. Thus we have something like 5 / 112

7. Causality We know We know that the state of Fuel? and the state of Clean Spark Plugs? have a causal impact on the state of Start?. Thus we have something like Start Clean Spark Plugs Fuel 5 / 112

9. Causal Structure - Judea Perl (1988) Definition A causal structure of a set of variables V is a directed acyclic graph (DAG) in which each node corresponds to a distinct element of V , and each edge represents direct functional relationship among the corresponding variables. Observation Causal Structure A precise specification of how each variable is influenced by its parents in the DAG. 7 / 112

10. Causal Structure - Judea Perl (1988) Definition A causal structure of a set of variables V is a directed acyclic graph (DAG) in which each node corresponds to a distinct element of V , and each edge represents direct functional relationship among the corresponding variables. Observation Causal Structure A precise specification of how each variable is influenced by its parents in the DAG. 7 / 112

11. Causal Model Deﬁnition A causal model is a pair M = D, ΘD consisting of a causal structure D and a set of parameters ΘD compatible with D. Thus The parameters ΘD assign a distribution xi = fi (pai, ui) ui ∼ p (ui) Where xi is a variable in the model D. pai are the parents of xi in D. ui is independent of any other u. 8 / 112

16. Example From the point of view of Statistics Formulation z = fZ (uZ ) x = fX (z, uX ) y = fY (x, uY ) 9 / 112

20. Now add an observation x0 From the point of view of Statistics Formulation after blocking information z = fZ (uZ ) x = x0 y = fY (x, uY ) 10 / 112

25. Causal Networks Deﬁnition A causal network consists of a set of variables and a set of directed links (also called edges) between variables. Thus In order to analyze a causal network is necessary to analyze: Causal Chains Common Causes Common Eﬀects 12 / 112

32. Causal Chains This conﬁguration is a “causal chain” What about the Joint Distribution? We have by the Chain Rule P (X, Y , Z) = P (X) P (Y |X) P (Z|Y , X) (1) 14 / 112

33. Causal Chains This conﬁguration is a “causal chain” What about the Joint Distribution? We have by the Chain Rule P (X, Y , Z) = P (X) P (Y |X) P (Z|Y , X) (1) 14 / 112

34. Propagation of Information Given no information about Y Information can propagate from X to Z. Thus The natural question is What does happen if Y = y for some value y? 15 / 112

35. Propagation of Information Given no information about Y Information can propagate from X to Z. Thus The natural question is What does happen if Y = y for some value y? 15 / 112

36. Thus, we have that Blocking Propagation of Information 16 / 112

37. Using Our Probabilities Then Z is independent of X given a Y = y And making the assumption that once an event happens P (Z|X, Y = y) = P (Z|Y = y) (2) YES!!! Evidence along the chain “blocks” the inﬂuence. 17 / 112

38. Using Our Probabilities Then Z is independent of X given a Y = y And making the assumption that once an event happens P (Z|X, Y = y) = P (Z|Y = y) (2) YES!!! Evidence along the chain “blocks” the inﬂuence. 17 / 112

39. Thus Something Notable Knowing that X has occurred does not make any diﬀerence to our beliefs about Z if we already know that Y has occurred. Thus conditional independencies can be written IP (Z, X|Y = y) (3) 18 / 112

40. Thus Something Notable Knowing that X has occurred does not make any diﬀerence to our beliefs about Z if we already know that Y has occurred. Thus conditional independencies can be written IP (Z, X|Y = y) (3) 18 / 112

41. Therefore The Joint Probability is equal to P (X, Y = y, Z) = P (X) P (Y = y|X) P (Z|Y = y) (4) 19 / 112

42. Thus Is X independent of Z given Y = y? P (Z|X, Y = y) = P (X, Y = y, Z) P (X, Y = y) = P (X) P (Y = y|X) P (Z|Y = y) P (X) P (Y = y|X) = P (Z|Y = y) 20 / 112

46. Common Causes Another basic conﬁguration: two eﬀects of the same cause Thus P (X, Y = y, Z) = P (X) P (Y = y|X) P (Z|X, Y = y) P(Z|Y =y) (5) 22 / 112

47. Common Causes Another basic conﬁguration: two eﬀects of the same cause Thus P (X, Y = y, Z) = P (X) P (Y = y|X) P (Z|X, Y = y) P(Z|Y =y) (5) 22 / 112

48. Common Causes What happened if X is independent of Z given Y = y? P (Z|X, Y = y) = P (X, Y = y, Z) P (X, Y = y) = P (X) P (Y = y|X) P (Z|Y = y) P (X) P (Y = y|X) = P (Z|Y = y) YES!!! Evidence on the top of the chain “blocks” the inﬂuence between X and Z. 23 / 112

51. Thus It gives rise to the same conditional independent structure as chains IP (Z, X|Y = y) (6) i.e. if we already know about Y , then an additional information about X will not tell us anything new about Z. 24 / 112

52. Thus It gives rise to the same conditional independent structure as chains IP (Z, X|Y = y) (6) i.e. if we already know about Y , then an additional information about X will not tell us anything new about Z. 24 / 112

54. Common Effect Last configuration: two causes of one effect (v-structures) Are X and Z independent if we do not have information about Y ? Yes!!! Because the ballgame and the rain can cause traffic, but they are not correlated. 26 / 112

55. Common Effect Last configuration: two causes of one effect (v-structures) Are X and Z independent if we do not have information about Y ? Yes!!! Because the ballgame and the rain can cause traffic, but they are not correlated. 26 / 112

60. Common Eﬀects Are X and Z independent given Y = y? No!!! Because seeing traﬃc puts the rain and the ballgame in competition as explanation!!! Why? P (X, Z|Y = y) = P (X, Z, Y = y) P (Y = y) = P (X|Z, Y = y) P (Z|Y = y) P (Y = y) P (Y = y) = P (X|Z, Y = y) P (Z|Y = y) 28 / 112

64. The General Case Backwards from the other cases Observing an eﬀect activates inﬂuence between possible causes. Fact Any complex example can be analyzed using these three canonical cases Question In a given Bayesian Network, Are two variables independent (given evidence)? Solution Analyze Graph Deeply!!! 29 / 112

69. Analyze the Graph Deﬁnition 2.1 Let G = (V , E) be a DAG, where V is a set of random variables. We say that, based on the Markov condition, G entails conditional independence: IP(A, B|C) for A, B, C ⊆ V if IP(A, B|C) holds for every P ∈ PG , where PG is the set of all probability distributions P such that (G, P) satisﬁes the Markov condition. Thus We also say the Markov condition entails the conditional independence for G and that the conditional independence is in G. 31 / 112

72. Analyze the Graph Question In a given Bayesian Network, Are two variables independent (Given evidence)? Solution Analyze Graph Deeply!!! 32 / 112

73. Analyze the Graph Question In a given Bayesian Network, Are two variables independent (Given evidence)? Solution Analyze Graph Deeply!!! 32 / 112

75. Examples of Entailed Conditional independence Example G: Graduate Program Quality. F: First Job Quality. B: Number of Publications. C: Number of Citations. F is given some evidence!!! G F B C 34 / 112

76. Examples of Entailed Conditional independence Example G: Graduate Program Quality. F: First Job Quality. B: Number of Publications. C: Number of Citations. F is given some evidence!!! G F B C 34 / 112

81. Finally We have Ip (C, G|F) (7) 36 / 112

82. D-Separation ≈ Conditional independence F and G are given as evidence Example C and G are d-separated by A, F in the DAG in G A F B C 37 / 112

84. Basic Deﬁnitions Deﬁnition (Undirected Paths) A path between two sets of nodes X and Y is any sequence of nodes between a member of X and a member of Y such that every adjacent pair of nodes is connected by an edge (regardless of direction) and no node appears in the sequence twice. 39 / 112

85. Example An example 40 / 112

86. Example Another one 41 / 112

87. Thus Given a path in G = (V , E) There are the edges connecting [X1, X2, ..., Xk]. Therefore Given the directed edge X → Y , we say the tail of the edge is at X and the head of the edge is Y . 42 / 112

90. Basic Classiﬁcations of Meetings Head-to-Tail A path X → Y → Z is a head-to-tail meeting, the edges meet head- to-tail at Y , and Y is a head-to-tail node on the path. Tail-to-Tail A path X ← Y → Z is a tail-to-tail meeting, the edges meet tail-to-tail at Z, and Z is a tail-to-tail node on the path. Head-to-Head A path X → Y ← Z is a head-to-head meeting, the edges meet head-to-head at Y , and Y is a head-to-head node on the path. 43 / 112

93. Examples Head-to-Tail 44 / 112

94. Examples Tail-to-Tail 45 / 112

95. Examples Head-to-Head 46 / 112

96. Basic Classiﬁcations of Meetings Finally A path (undirected) X − Z − Y , such that X and Y are not adjacent, is an uncoupled meeting. 47 / 112

98. Blocking Information ≈ Conditional Independence Deﬁnition 2.2 Deﬁnition 2.2 LetG = (V , E) be a DAG, A ⊆ V , X and Y be distinct nodes in V − A, and ρ be a path between X and Y . Then ρ is blocked by A if one of the following holds: 1 There is a node Z ∈ A on the path ρ, and the edges incident to Z on ρ meet head-to-tail at Z. 2 There is a node Z ∈ A on the path ρ, and the edges incident to Z on ρ, meet tail-to-tail at Z. 3 There is a node Z, such that Z and all of Z’s descendent’s are not in A, on the chain ρ, and the edges incident to Z on ρ meet head-to-head at Z. 49 / 112

103. Example We have that the path [Y , X, Z, S] is blocked by {X} and {Z} Because the edges on the chain incident to X meet tail-to-tail at X. 50 / 112

104. Example We have that the path [W , Y , R, Z, S] is blocked by ∅ Because R /∈ ∅ and T /∈ ∅ and the edges on the chain incident to R meet head-to-head at R . 51 / 112

106. Definition of D-Separation Definition 2.3 Let G = (V , E) be a DAG, A ⊆ V , and X and Y be distinct nodes in V − A. We say X and Y are D-Separated by A in G if every path between X and Y is blocked by A. Definition 2.4 Let G = (V , E) be a DAG, and A, B, and C be mutually disjoint subsets of V . We say A and B are d-separated by C in G if for every X ∈ A and Y ∈ B, X and Y are D-Separated by C. We write IG(A, B|C) or A ⊥⊥ B|C (8) If C = ∅, we write only IG(A, B) or A ⊥⊥ B. 53 / 112

109. Example X and T are D-Separated by {Y , Z} Because the chain [X, Y , R, T] is blocked at Y . 54 / 112

110. Example X and T are D-Separated by {Y , Z} Because the chain [X, Z, S, R, T] is blocked at Z, S. 55 / 112

111. D-Separation ⇒ Independence D-Separation Theorem Let P be a probability distribution of the variables in V and G = (V , E) be a DAG. Then (G, P) satisﬁes the Markov condition if and only if for every three mutually disjoint subsets A, B, C ⊆ V , whenever A and B are D-Separated by C, A and B are conditionally independent in P given C. That is, (G, P) satisﬁes the Markov condition if and only if IG (A, B|C) ⇒ IP (A, B|C) (9) 56 / 112

114. Proof The proof that, if (G, P) satisﬁes the Markov condition Then, each D-Separation implies the corresponding conditional independence is quite lengthy and can be found in [Verma and Pearl, 1990] and in [Neapolitan, 1990]. Then, we will only prove the other direction Suppose each D-Separation implies a conditional independence. Thus, the following implication holds IG (A, B|C) ⇒ IP (A, B|C) (10) 57 / 112

117. Proof Something Notable It is not hard to see that a node’s parents D-Separate the node from all its non-descendent’s that are not its parents. This is If we denote the sets of parents and non-descendent’s of X by PAX and NDX respectively, we have IG ({X} , NDX − PAX |PAX ) (11) Thus IP ({X} , NDX − PAX |PAX ) (12) 58 / 112

120. Proof This states the same than IP ({X} , NDX |PAX ) (13) Meaning The Markov condition is satisﬁed. 59 / 112

121. Proof This states the same than IP ({X} , NDX |PAX ) (13) Meaning The Markov condition is satisﬁed. 59 / 112

122. Every Entailed Conditional Independence is Identiﬁed by D-Separation Lemma 2.2 Any conditional independence entailed by a DAG, based on the Markov condition, is equivalent to a conditional independence among disjoint sets of random variables. Theorem 2.1 Based on the Markov condition, a DAG G entails all and only those conditional independences that are identiﬁed by D-Separations in G. 60 / 112

123. Every Entailed Conditional Independence is Identiﬁed by D-Separation Lemma 2.2 Any conditional independence entailed by a DAG, based on the Markov condition, is equivalent to a conditional independence among disjoint sets of random variables. Theorem 2.1 Based on the Markov condition, a DAG G entails all and only those conditional independences that are identiﬁed by D-Separations in G. 60 / 112

125. Now We would like to find the D-Separations Since d-separations entail conditional independencies. We want an efficient algorithm For determining whether two sets are D-Separated by another set. For This, we need to build an algorithm One that can find all D-Separated nodes from one set of nodes by another. 62 / 112

128. How? To accomplish this We will ﬁrst ﬁnd every node X such that there is at least one active path given A between X and a node in D. Something like 63 / 112

129. How? To accomplish this We will ﬁrst ﬁnd every node X such that there is at least one active path given A between X and a node in D. Something like 63 / 112

130. We solve the following problem Suppose we have a directed graph We say that certain edges cannot appear consecutively in our paths of interest. Thus, we identify certain pair of edges (u → v, v → w) As legal and the rest as illegal!! Legal? We call a path legal if it does not contain any illegal ordered pairs of edges. We say Y is reachable from x if there is a legal path from x to y . 64 / 112

134. Thus We can ﬁnd the set R of all nodes reachable from x as follows Any node V such that the edge x → v exists is reachable. Then We label such edge with 1. Next for each such v We check all unlabeled edges v → w and see if (x → v, v → w) is a legal pair. 65 / 112

137. Then We label each such edge with a 2 And keep going!!! Similar to a Breadth-First Graph Here, we are visiting links rather than nodes. 66 / 112

138. Then We label each such edge with a 2 And keep going!!! Similar to a Breadth-First Graph Here, we are visiting links rather than nodes. 66 / 112

139. What do we want? Identifying the set of nodes D that are The one that are D-Separated from B by A in a DAG G = (V , E). For this We need to ﬁnd the set R such that y ∈ R ⇐⇒ Either y ∈ B. There is at least one active chain given A between y and a node in B. 67 / 112

144. Then If there is an active path ρ between node X and some other node Then every 3-node sub-path u − v − w (u = w) of ρ has the following property. Either u − v − w is not head-to-head at v and v is not in A. Or u − v − w is a head-to-head at v and v is a or has a descendant in A. 68 / 112

147. The Final Legal Rule The algorithm ﬁnd-reachable-nodes uses the RULE Find if (u → v, v → w) is legal in G . The pair (u → v, v → w) is legal if and only if u = w (14) And one of the following holds 1 (u − v − w) is not head-to-head in G and inA [v] is false. 2 (u − v − w) is head-to-head in G and descendent [v] is true. 69 / 112

152. Example Reachable nodes from X when A = ∅, thus inA [v] is false and descendent [v] is false for all v ∈ V Therefore Only the rule 1 is applicable. 71 / 112

153. Example Reachable nodes from X when A = ∅, thus inA [v] is false and descendent [v] is false for all v ∈ V Therefore Only the rule 1 is applicable. 71 / 112

154. Example Labeling edges to 1 and shaded nodes are in R 72 / 112

155. Example Labeling edges to 1 and shaded nodes are in R 1 1 1 73 / 112

156. Example Labeling edges to 2 and shaded nodes are in R 1 1 1 2 2 74 / 112

157. Example Labeling edges to 3 and shaded nodes are in R 3 3 31 1 1 2 2 75 / 112

158. Example Labeling edges to 4 and shaded nodes are in R 3 3 4 4 31 1 1 2 2 76 / 112

159. Example Labeling edges to 5 and shaded nodes are in R 3 3 4 4 31 1 1 2 2 5 77 / 112

161. Algorithm to Finding Reachability ﬁnd-reachable-nodes(G, set of nodes B, set of nodes &R) Input:G = (V,E), subset B ⊂ V and a RULE to ﬁnd if two consecutive edges are legal Output: R ⊂ V of all nodes reachable from B 1. for each x ∈ B 2. add x to R 3. for (each v such that x → v ∈ E) 5. add v to R 6. label x → v with 1 7. i = 1 8. found =true 79 / 112

165. Continuation The following steps 9. while (found) 10. found =false 11. for (each v such that u → v labeled i) 12. for (each unlabeled edge v → w 13. such (u → v, v → w) is legal) 14. add w to R 15. label v → w with i + 1 16. found =true 17. i = i + 1 80 / 112

169. Complexity of ﬁnd-reachable-nodes We have that Let n be the number of nodes and m be the number of edges. Something Notable In the worst case, each of the nodes can be reached from n entry points. Thus Each time a node is reached, an edge emanating from it may need to be re-examined. 82 / 112

172. Complexity of ﬁnd-reachable-nodes Then Then, in the worst case each edge may be examined n times Thus, the complexity W (m, n) = Θ (mn) (15) 83 / 112

173. Complexity of ﬁnd-reachable-nodes Then Then, in the worst case each edge may be examined n times Thus, the complexity W (m, n) = Θ (mn) (15) 83 / 112

175. Algorithm for D-separation Find-D- Separations(DAG G = (V,E) , set of nodes A,B, set of nodes D) Input: G = (V,E)and two disjoint subsets A,B ⊂ V Output: D ⊂ V containing all nodes D- Separated from every node in B by A. That is IG (B,D|A) holds and no superset D has this property. 1. for each v ∈ V 2. if (v ∈ A) 3. inA [v] = true 4. else 5. inA [v] = false 6. if (v is or has a descendent in A) 7. descendent [v] = true 8. else 9. descendent [v] = false 10. E = E ∪ {u → v|v → u ∈ E} 11. G = (V,E’) 12. Run the algorithm: ﬁnd-reachable-nodes(G , B,R) Note B ⊆ R 13. return D=V − (A ∪ R) 85 / 112

183. Observation about descendent [v] We can implement the construction of descendent [v] as follow Initially set descendent [v] = true for all nodes in A. Then Then follow the incoming edges in A to their parents, their parents’ parents, and so on. Thus We set descendent [v] = true for each node found along the way. 86 / 112

186. Observation about E’ The RULE about legal and E’ The E’ is necessary because using only the RULE on E will no get us all the active paths. For example 87 / 112

187. Observation about E’ The RULE about legal and E’ The E’ is necessary because using only the RULE on E will no get us all the active paths. For example 1 1 2 2 87 / 112

188. Thus Something Notable Given A is the only node in A and X → T is the only edge in B, the edges in that DAG are numbered according to the method. Then The active chain X → A ← Z ← T ← Y is missed because the edge T → Z is already numbered by the time the chain A ← Z ← T is investigated. But If we use the set of edges E’. 88 / 112

191. Thus Once, we add the extra edges, we get 1 1 2 2 3 4 89 / 112

193. Complexity Please take a look at page 81 at “Learning Bayesian Networks” by Richard E. Neapolitan. For the analysis of the algorithm for m edges and n nodes Θ (m) with m ≥ n. (16) 91 / 112

194. Complexity Please take a look at page 81 at “Learning Bayesian Networks” by Richard E. Neapolitan. For the analysis of the algorithm for m edges and n nodes Θ (m) with m ≥ n. (16) 91 / 112

195. The D-Separation Algorithm works Theorem 2.2 The set D contains all and only nodes D-Separated from every node in B by A. That is, we have IG (B,D|A) and no superset of D has this property. Proof The set R determined by the algorithm contains All nodes in B. All nodes reachable from B via a legal path in G . 92 / 112

196. The D-Separation Algorithm works Theorem 2.2 The set D contains all and only nodes D-Separated from every node in B by A. That is, we have IG (B,D|A) and no superset of D has this property. Proof The set R determined by the algorithm contains All nodes in B. All nodes reachable from B via a legal path in G . 92 / 112

197. Proof For any two nodes x ∈ B and y /∈ A ∪ B The path x − ... − y is active in G if and only if the path x → ... → y is legal in G . Thus Thus R contains the nodes in B plus all and only those nodes that have active paths between them and a node in B. By the deﬁnition of D-Separation A node is D-Separated from every node in B in A if the node is not in A ∪ B and there is not active path between the node and a node in B. 93 / 112

200. Proof Thus D=V-(A ∪ R) is the set of all nodes D-Separated from every node in B by A. 94 / 112

202. Example B = {x} and A = {s, v} 96 / 112

203. Example B = {x} and A = {s, v} and the ﬁrst part of the reachability algorithm 1 1 1 1 97 / 112

204. Example Remember that Legality is in G 1 1 1 2 1 2 2 2 98 / 112

205. Example Now, we have that 1 1 1 2 1 32 2 2 99 / 112

206. Example D − {A ∪ R} = {q} 1 1 1 2 1 3 4 2 2 2 100 / 112

208. Application Something Notable In general, the inference problem in Bayesian networks is to determine P (B|A), where A and B are two sets of variables. Thus We can use the D-Separation for that. 102 / 112

209. Application Something Notable In general, the inference problem in Bayesian networks is to determine P (B|A), where A and B are two sets of variables. Thus We can use the D-Separation for that. 102 / 112

210. Example Given the following the DAG G 103 / 112

211. Example We generate G 104 / 112

212. Where The extra nodes represent The probabilities in the interval [0, 1] and representing P (X = x) Creating a set of P be the set of auxiliary parent nodes Thus, if we want to determine P (B|A) in G, we can use the algorithm for D-Separation to ﬁnd D. Such that IG (B,D|A) (17) And no superset of D has this property, then take D∩P. 105 / 112

215. For example We want P (f ) To determine it we need that all and only the values of PH , PB , PL , and PF Because IG ({F} , {PX } |∅) (18) Thus PX is the only auxiliary parent variable D-Separated from {F} by the empty set. 106 / 112

218. For example We want P (f |b) To determine it we need that all and only the values of PH , PL , and PF when separation set is {B} Then IG ({F} , {PX , PB} | {B}) (19) Thus PX is the only auxiliary parent variable D-Separated from {F} by the empty set. 107 / 112

222. Remarks Bayes Networks can reflect the true causal patterns Often simpler (nodes have fewer parents). Often easier to think about. Often easier to elicit from experts. Something Notable Sometimes no causal net exists over the domain. For example, consider the variables Traffic and Drips. Arrows reflect correlation not causation. 109 / 112

228. Remarks What do the arrows really mean? Topologies may happen to encode causal structure. Topologies are only guaranteed to encode conditional independence! 110 / 112

229. Remarks What do the arrows really mean? Topologies may happen to encode causal structure. Topologies are only guaranteed to encode conditional independence! 110 / 112

230. Example Example Burgalary Earthquake Alarm John Calls Mary Calls Add more stuﬀ Given that some information is not being encoded into the network: We have to add more edges to the graph. 111 / 112

231. Example Example Burgalary Earthquake Alarm John Calls Mary Calls Add more stuﬀ Given that some information is not being encoded into the network: We have to add more edges to the graph. 111 / 112

232. Example Thus Adding edges allows to make diﬀerent conditional independence assumptions. New Network 112 / 112

233. Example Thus Adding edges allows to make diﬀerent conditional independence assumptions. New Network Burgalary Earthquake Alarm John Calls Mary Calls 112 / 112

Artificial Intelligence 06.2 More on Causality Bayesian Networks

More Related Content

What's hot (20)

Viewers also liked (17)

Similar to Artificial Intelligence 06.2 More on Causality Bayesian Networks (20)

More from Andres Mendez-Vazquez (20)

Recently uploaded (20)

Artificial Intelligence 06.2 More on Causality Bayesian Networks