35
Deep Learning Finding Deep Structures in Data Chris Orwa 3 rd March 2016

Deep learning

Embed Size (px)

Citation preview

Page 1: Deep learning

Deep LearningFinding Deep Structures in Data

Chris Orwa3rd March 2016

Page 2: Deep learning

What is Deep Learning?

Page 3: Deep learning

“A branch of machine learning based on a set of algorithms that attempt to model high-level abstractions in data” - Wikipedia

Page 4: Deep learning

Tenets of Deep Learning• Multiple Processing Layers• Model data in high abstraction• Learn from abstraction

Page 5: Deep learning

Our Processing Layers

•Principal Component Analysis• Mantel’s Test• Time Series• Group Theory

Page 6: Deep learning

Principal Component Analysis • Allow data representation in alternative co-ordinate system. • Provides for dimensionality reduction• Easy to visualize• Emphasize variation in data

Page 7: Deep learning

Run PCA on Questionnaire

Page 8: Deep learning

Inconsistency!

Page 9: Deep learning

Track Changes

Page 10: Deep learning

Inconsistency!

Page 11: Deep learning

How can I better capture the changes?

Page 12: Deep learning

Mantel’s Permutation Test

Page 13: Deep learning

Mantel’s Test

• Conceived by Nathan Mantel in 1967 • Used to track cancer epidemics• Test correlation between two matrices• Supported in R by APE package

Page 14: Deep learning

Modify Code

Page 15: Deep learning

Modify Code Cont’d

Page 16: Deep learning

What Did I Get?

Page 17: Deep learning

Same Test on Twitter Data #EastleighBlast

Page 18: Deep learning

Matrix Representation

Page 19: Deep learning
Page 20: Deep learning
Page 21: Deep learning

Same Test on Forex Data CADUSD,EURGBP, USDJPY

Page 22: Deep learning
Page 23: Deep learning
Page 24: Deep learning
Page 25: Deep learning
Page 26: Deep learning

Observations!

Page 27: Deep learning

There are Two Types of Data• Stable Structure: Does not vary with variation.• Unstable Structure: Varies with variation.

Page 28: Deep learning

Application!

Page 29: Deep learning

Scenario I: Stable Structures

Page 30: Deep learning

Lend Themselves to Two Important Fields

• Spatial Principal Component Analysis

• Group Theory

Page 31: Deep learning

Library(adegenet)

Page 32: Deep learning

Scenario II: Unstable Structures

Page 33: Deep learning

Two Application Areas

•Seasonal Market-Basket Analysis

• Tracking Online Conversations

Page 34: Deep learning

Sentiment Analysis

Page 35: Deep learning

Thank You!blackorwa.com

@blackorwa