[关闭]
@nanmeng 2016-06-16T13:48:38.000000Z 字数 795 阅读 808

Probabilistic Graphical Models(Stanford) - 5

notes Probabilistic_Graphical_Models

Week3 Knowledge Engineering

1. Knowledge Engineering

examples:

Template-based Specific
image segmentation medical diagnosis
fault diagnosis fault diagnosis
small number of variable type large number of "unique" variables

PGM3_1

PGM3_2
In the case that the data are not fully labeled, it turns out that sometimes you cannot train discrimitive models but you can train generative models

Variable Types

Structure

PGM3_4

Extending the Conversation

PGM3_5

Parameters

PGM3_6
PGM3_7

Iterative Refinement

PGM3_8

Questions

PGM3_9

2. Overview: Conditional Probability Queries

PGM3_10

NP-hardness

PGM3_11

So one can see even find a approximate answer is still NP-hard...

PGM3_12

MAP Inference(Maximum a Posterior)


MAP is NP-hard
PGM3_13

Variable Elimination Algorithm

PGM3_14

--

添加新批注
在作者公开此批注前,只有你和作者可见。
回复批注