Tensorflow London: Tensorflow and Graph Recommender Networks by Yaz Santissi

of 36
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information Report
Category:

Technology

Published:

Views: 0 | Pages: 36

Extension: PDF | Download: 0

Share
Description
Speaker: Yaz Santissi, GDG Cloud Title: Tensorflow and Graph Recommender Networks Abstract: Recommender systems are a core part of the web landscape - from social media, e-commerce, transport, search, and content networks to name just a few. But, compared to other areas such as image recognition, success in Tensorflow for recommender systems has lagged behind. Yaz will provide an overview of the latest research in the field, how it compares to traditional approaches, and discuss the implications for future developments. Bio: Despite working in cloud and machine learning, Yaz believes creativity or power is still not measured in terms of numbers of GPUs or scalable servers. While appreciating what they can do, he remains a steadfast fan of human learning. In fact, he organises the largest number of TensorFlow events in the world. Usually, in a hack format, participants not only share what they have learned, they also draw inspiration from each other, build projects together, and experiment with the process of coding collaboratively itself in order to nurture that sometimes elusive developer flow.
Transcript
  • 1. TensorFlow + Graph Recommender Networks Yaz Santissi @yazdotai
  • 2. Machine Learning AI
  • 3. It depends ?
  • 4. 5
  • 5. 7
  • 6. 8
  • 7. Movie Lens ( 1 million ) Reddit ( 11 million )
  • 8. Is Graph Data Not Important ?
  • 9. 1 billion hours per day 80 %
  • 10. $140 billion 10x
  • 11. 44 trillion GB by 2020 300x than in 2005 90% in past two years 1.7MB per human per sec
  • 12. So why has TensorFlow not taken over graph data?
  • 13. Graph Data does not need TensorFlow to be useful
  • 14. Limits of Feature Engineering
  • 15. Can we not port established TF algorithms to graph data?
  • 16. Euclidean mathematics domain
  • 17. - Two points determine a unique line - Any terminated line may be extended indefinitely - A circle may be drawn with any given point as center and any given radius - All right angles are equal
  • 18. - Shortest distance bet two points will always be a straight line - Interior angles of a triangle add up to 180 degrees - Two parallel lines will never cross
  • 19. Deep Walk + Node2Vec
  • 20. GraphSAGE
  • 21. 3 billion nodes 18 billion edges
  • 22. Future Directions
  • 23. Yaz Santissi @yazdotai Thank You https://github.com/yazdotai/ graph-networks
  • Recommended
    View more...
    We Need Your Support
    Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

    Thanks to everyone for your continued support.

    No, Thanks
    SAVE OUR EARTH

    We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

    More details...

    Sign Now!

    We are very appreciated for your Prompt Action!

    x