Lecture 6

# I p gi dp s i p lgp j l s p h g j

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: D) D φC (C )φD (D, C ) C I φH (H, G, J ) τ3(G, S ) φL (L, G) φJ (J, L, S ) = C φL (L, G) φ( G, I, D)τ1(D) φS (S, I )φI (I )τ2(G, I ) φH (H, G, J ) H G S = φC (C )φD (D, C )φI (I )φG (G, I, D)φS (S, I )φL(L, G)φJ (J, L, S )φH (H, G, J ) = φC (C )φD (D, C ) C D I H φL (L, G) φJ (J, L, S ) L P (C )P (D|C )P (I )P (G|I, D)P (S |I )P (L|G)P (J |L, S )P (H |G, J ) = φ( G, I, D) D φS (S, I )φI (I ) φH (H, G, J ) φL (L, G) C = φS (S, I )φI (I ) I τ2 (G,I ) P (C, D, I, G, S, L, J, H ) S φH (H, G, J ) H G S = • Key idea 2: use (non-serial) dynamic programming to cache shared subexpressions. L φL (L, G) G φJ (J, L, S ) = • Key idea 1: push sum inside products. P (J ) = S τ1 (D) Letter Happy φJ (J, L, S ) L SAT τ6 (J, L) = L τ7 (J ) Different ordering Dealing with evidence: method 1 Coherence D H C L Difficulty G I S φG (G, I, D)φL(L, )φH (H, G, J ) φI (I )φS (S, I ) φJ (J, L, S ) φD (D, C ) P (J ) = Grade τ1 (I,D,L,J,H ) = φD (D, C ) D C φJ (J, L, S ) H L S φD (D, C ) D C φJ (J, L, S )τ2(D, L, S, J, H ) H L φD (D, C ) D C τ3(D, L, J, H ) H L τ4 (D,J,H ) = φD (D, C ) D C τ4 (D, J, H ) C,D,G,L,S φD (D, C )τ5(D, J ) D • The denominator is P (e) = P (I = 1, H = 0). C τ6 (D,J ) = Job H τ5 (D,J ) = Happy • We can instantiate observed variables to their observed value: P (J, I = 1, H = 0) ∝ P (J, I = 1, H = 0) P (J |I = 1, H = 0) = P (I = 1, H = 0) = P (C, D, I = 1, G, S, L, J, H = 0) S τ3 (D,L,J,H ) = SAT Letter φI (I )φS (S, I )τ1(I, D, L, J, H ) I τ2 (D,L,S,J,H ) = Intelligence • For Markov networks, the denominator is P (e) × Z . τ6 (D, J ) D τ7 (J ) Dealing with evidence: method 2 • We can associate a local evidence potential with every node, and set φi(Xi) = δ (Xi, x∗) if Xi is observed to have value x∗, and i i φi(Xi) = 1 otherwise: P (X1:n|ev ) ∝ P (X1:n) P (evi|Xi) i • e.g., P (J |I = 1, H = 0) ∝ P (C, D, I, G, S, L, J, H )δI (I, 1)δH (H, 0) C,D,I,G,S,L,J,H...
View Full Document

Ask a homework question - tutors are online