Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Deep Learning in Deep Space

3,715 views

Published on

Dr. Brian Mac Namee
Insight@DCU Machine Learning Workshop
Dublin City University, Dublin, Ireland
April 27, 2017

https://telecombcn-dl.github.io/dlmm-2017-dcu/

Published in: Data & Analytics
  • Be the first to comment

Deep Learning in Deep Space

  1. 1. Deep Learning in Deep Space Applica(ons of Machine Learning on Astronomical Data Dr. Brian Mac Namee brian.macnamee@ucd.ie @brianmacnamee
  2. 2. Tycho Brahe (1546 –1601)
  3. 3. Big sky survey projects are amongst the biuggest data sources available Sloan Digital Sky Survey (www.sdss.org)
  4. 4. Sloan Digital Sky Survey (www.sdss.org)
  5. 5. Sloan Digital Sky Survey (www.sdss.org)
  6. 6. The SDSS telescopes collect over 200GB of data every night Over 200 million galaxies have been photographed Data has become a huge boPle neck in modern astronomy Future telescopes will collect even more data Sloan Digital Sky Survey (www.sdss.org)
  7. 7. Data Collec(on Data Processing Science
  8. 8. Data Collec(on Data Processing Science Machine Learning * Connec(on from data to Brahe to Newton stolen from Prof. James Gleeson (UL)
  9. 9. Data Collec(on Data Processing Science Machine Learning * Connec(on from data to Brahe to Newton stolen from Prof. James Gleeson (UL) (Machine Learning) Correla(on or Causa(on (Sta(s(cs)
  10. 10. Data Collec(on Data Processing Science Machine Learning * Connec(on from data to Brahe to Newton stolen from Prof. James Gleeson (UL) (Machine Learning) Correla(on or Causa(on (Sta(s(cs)
  11. 11. Ellipse Le0 Spiral Right Spiral Fundamentals of Machine Learning for Predic(ve Data Analy(cs John D. Kelleher, Brian Mac Namee, Aoife D'Arcy MIT Press www.machinelearningbook.com
  12. 12. Support Vector Machine Performance Fundamentals of Machine Learning for Predic(ve Data Analy(cs John D. Kelleher, Brian Mac Namee, Aoife D'Arcy MIT Press www.machinelearningbook.com Overall accuracy: ~88%
  13. 13. Support Vector Machine Performance Fundamentals of Machine Learning for Predic(ve Data Analy(cs John D. Kelleher, Brian Mac Namee, Aoife D'Arcy MIT Press www.machinelearningbook.com Convolu(onal Neural Network Performance Overall accuracy: ~88% Overall accuracy: ~98% "Rota(on-invariant convolu(onal neural networks for galaxy morphology predic(on.", Dieleman, Sander, Kyle W. WilleP, and Joni Dambre. Monthly no)ces of the royal astronomical society 450, no. 2 (2015): 1441-1459.
  14. 14. Galaxy Catalogue (hPp://www.zsolt-frei.net/galaxy_catalog.html) Data is the Fuel of Machine Learning
  15. 15. Galaxy Catalogue (hPp://www.zsolt-frei.net/galaxy_catalog.html) Data is the Fuel of Machine Learning Labelled data is the real key
  16. 16. Galaxy Zoo (www.galaxyzoo.org) Crowdsourcing Labelled Data
  17. 17. Galaxy Zoo (www.galaxyzoo.org) Crowdsourcing Labelled Data Almost 1,000,000 galaxies labelled Over 150,000 users Millions of individual labels
  18. 18. Other uses of machine learning in astronomy: – Object classifica(on (stars, galaxies, quasars, ...) – Es(ma(ng redshil – Es(ma(ng other parameters of sky objects – Forecas(ng solar flare ac(vity Kremer, Jan, Kristoffer Stensbo-Smidt, Fabian Gieseke, Kim Steenstrup Pedersen, and Chris(an Igel. "Big Universe, Big Data: Machine Learning and Image Analysis for Astronomy." IEEE Intelligent Systems 32, no. 2 (2017): 16-22. Ball, Nicholas M., and Robert J. Brunner. "Data mining and machine learning in astronomy." Interna)onal Journal of Modern Physics D 19, no. 07 (2010): 1049-1106.
  19. 19. European Space Agency (www.galaxyzoo.org) Earth Observa(on
  20. 20. Earth Observa(on Terrain classifica(on is a standard task Convolu(onal neural networks are leading approach for this task Accuracies of >90% possible Classifica(on and segmenta(on of satellite orthoimagery using convolu(onal neural networks, Längkvist, Mar(n, Andrey Kiselev, Marjan Alirezaie, and Amy Lousi, Remote Sensing 8, no. 4 (2016): 329.
  21. 21. Other uses of machine learning in earth observa(on: – Offers poten(al to use mul(-spectral images – Terrain classifica(on – Target recogni(on – Seman(c feature extrac(on – Scene understanding
  22. 22. Thank you Ques<ons? Dr. Brian Mac Namee brian.macnamee@ucd.ie @brianmacnamee www.bigskyearth.eu www.machinelearningbook.com

×