Embed
Email

Markov Logic and Deep Networks

Document Sample
Markov Logic and Deep Networks
Shared by: HC11120422384
Categories
Tags
Stats
views:
0
posted:
12/4/2011
language:
English
pages:
6
Markov Logic and

Deep Networks

Pedro Domingos

Dept. of Computer Science & Eng.

University of Washington

Markov Logic Networks

 Basic idea: Use first-order logic to

compactly specify large non-i.i.d. models

 MLN = Set of formulas with weights

 Formula = Feature template (Vars→Objects)

 E.g., HMM: State(+s,t) ^ State(+s',t+1)

Obs(+o,t) ^ State(+s,t)

1  

P( x)  exp   wi ni ( x) 

Z  i 

Weight of formula i No. of true instances of formula i in x

State of the Art in MLNs

 Many algorithms for learning and inference

 Inference: Millions of variables, billions of features

 Learning: Generative, discriminative, max-margin, etc.

 Best-performing solutions in many application areas

 Natural language, robot mapping, social networks,

computational biology, activity recognition, etc.

 Open-source software/Web site: Alchemy

alchemy.cs.washington.edu

 Book: Domingos & Lowd, Markov Logic,

Morgan & Claypool, 2009.

Deep Uses of MLNs

 Very large scale inference

 Defining architecture of deep networks

 Adding knowledge to deep networks

 Transition to natural language input

 Learning with many levels of hidden variables

MLNs for Deep Learning

 Basic idea: Use small amounts of knowledge

and large amounts of joint inference

to make up for lack of supervision

 Relational clustering:

 Cluster objects with similar relations to similar objects

 Cluster relations that hold between similar sets of objects

 Coreference resolution: Outperforms supervised

approaches on MUC and ACE benchmarks

 Semantic network extraction: Learns thousands of

concepts and relations from millions of Web triples

Example:

Unsupervised Semantic Parsing

 Goal: Read text and answer questions

 No supervision or annotation

 Input: Dependency parses of sentences

(Nodes→Unary predicates / Edges→Binary preds.)

 Outputs: Semantic parser and knowledge base

 Basic idea: Cluster expressions with similar

subexpressions

 Maps syntactic variants to common meaning

 Discovers its own meaning representation

 “Part of” lattice of clusters

 Applied to corpus of biomedical abstracts

 Three times more correct answers than next best


Related docs
Other docs by HC11120422384
No Slide Title
Views: 3  |  Downloads: 0
PowerPoint Presentation
Views: 6  |  Downloads: 0
Florida International University
Views: 0  |  Downloads: 0
Speakers
Views: 1  |  Downloads: 0
Hoja1
Views: 0  |  Downloads: 0
99SSP_Template
Views: 1  |  Downloads: 0
??? ???? ?????? ??????
Views: 0  |  Downloads: 0
TOWNSHIP OF MAHWAH PLANNING BOARD
Views: 0  |  Downloads: 0
?3? ?? ???
Views: 1  |  Downloads: 0
By registering with docstoc.com you agree to our
privacy policy

You are almost ready to download!

You are almost ready to download!