Presentation is loading. Please wait.

Presentation is loading. Please wait.

Learning Probabilistic Graphical Models Overview Learning Problems.

Similar presentations


Presentation on theme: "Learning Probabilistic Graphical Models Overview Learning Problems."— Presentation transcript:

1 Learning Probabilistic Graphical Models Overview Learning Problems

2 Learning Data Network dataset of instances D={d[1],...d[m]}
domain expert Declarative representation Network Learning elicitation

3 Learning Tasks Known structure Unknown structure Latent variables
Fully observable data Partly observable data

4 Known Structure, Complete Data
X1 X2 X1 X2 Inducer Initial network Y Y X1 X2 Y x10 x21 y0 x11 x20 y1 P(Y|X1,X2) X1 X2 y0 y1 x10 x20 1 x21 0.2 0.8 x11 0.1 0.9 0.02 0.98 Input Data

5 Unknown Structure, Complete Data
X1 X2 X1 X2 Inducer Initial network Y Y X1 X2 Y x10 x21 y0 x11 x20 y1 P(Y|X1,X2) X1 X2 y0 y1 x10 x20 1 x21 0.2 0.8 x11 0.1 0.9 0.02 0.98 Input Data

6 Known Structure, Incomplete Data
X1 X2 X1 X2 Inducer Initial network Y Y X1 X2 Y ? x21 y0 x11 x10 x20 y1 P(Y|X1,X2) X1 X2 y0 y1 x10 x20 1 x21 0.2 0.8 x11 0.1 0.9 0.02 0.98 Input Data

7 Unknown Structure, Incomplete Data
X1 X2 X1 X2 Inducer Initial network Y Y X1 X2 Y ? x21 y0 x11 x10 x20 y1 P(Y|X1,X2) X1 X2 y0 y1 x10 x20 1 x21 0.2 0.8 x11 0.1 0.9 0.02 0.98 Input Data

8 Latent Variables, Incomplete Data
H X1 X2 X1 X2 Inducer Initial network Y Y X1 X2 Y ? x21 y0 x11 x10 x20 y1 P(Y|X1,X2) X1 X2 y0 y1 x10 x20 1 x21 0.2 0.8 x11 0.1 0.9 0.02 0.98 Input Data

9 Learning Tasks: BNs Known structure Unknown structure Latent variables
Fully observable data Partly observable data

10 Learning Tasks: MNs Known structure Unknown structure Latent variables
Fully observable data Partly observable data

11 Target Tasks Probabilistic queries about new instances
Specific tasks (e.g., classification) Knowledge discovery Direct vs indirect dependencies Possibly directionality of edges Directionality of influence Hidden variables

12 Performance Metrics Distance of learned model to true distribution
Evaluate performance by how well the network estimates probability of new examples (“test data”) Task-specific accuracy on test set Distance of learned structure or parameters to true a ground truth model Often compare to (limited) prior knowledge

13 END END END


Download ppt "Learning Probabilistic Graphical Models Overview Learning Problems."

Similar presentations


Ads by Google