Upgrade to Pro — share decks privately, control downloads, hide ads and more …

MLretreat-2019.pdf

 MLretreat-2019.pdf

Ecdea9b9714877b86cee08458f085481?s=128

Tania Allard

June 26, 2019
Tweet

Transcript

  1. The places we'll go Tania Allard, PhD Developer Advocate @

    Microsoft Google Developer Expert ML/Tensorflow Industrial fellow Alan Turing Institute
  2. 2019... And we still see these headlines 2

  3. None
  4. Have deeply thought on how AI is embedded in your

    daily lives? How many of you
  5. None
  6. Let’s make AI boring

  7. Better call centres

  8. The vast amounts of data available The plethora of open

    source tools And even the number of open access journals and open data sets It’s an exciting time to be doing Machine Learning 8
  9. None
  10. Need to make a distinction 10 Machine automation - Previously

    observed outcomes Machine discovery - Search for novel or unknown outcomes Increased data and algorithmic complexity
  11. ML for scientific discovery

  12. Machine learning to identify pairwise interactions between specific IgE antibodies

    and their association with asthma 12 What Researchers applied network analysis and hierarchical clustering (HC) to explore the connectivity structure of component-specific IgEs and identified seven clusters of component-specific sensitisation. Cluster membership mapped closely to the structural homology of proteins and/or their biological source. https://doi.org/10.1371/journal.pmed.1002691
  13. Discovering new planets 13

  14. None
  15. Integrated ANN pipeline for biomarker discovery and validation in Alzheimer's

    disease 15 Segler, M., Preuss M., Waller M. Nature 2018 v555
  16. Moving from discovery 16

  17. Transitioning from lab to industry

  18. None
  19. Right framework Right production Right strategy

  20. None
  21. (ML) Research starts with good data

  22. What does this even mean? As the widespread of ML

    increases so does the evident lack of high-quality data
  23. None
  24. You are then living in the dumpster

  25. A quick glance Build models easily Scarce data and incomplete

    data Deploy to production Can only predict from what has learned before Scale Also sparsity scales
  26. https://hackernoon.com/the-ai-hierarchy-of-needs-18f111fcc007

  27. Issues we can tackle now Data unification Bringing data together

    into one unified data context Error detection While data cleaning has long been a research topic in academia, it often has been looked at as a theoretical logic problem
  28. Wonder where you can make the most impact? • In

    industry: ◦ Machine automation • In academia* • Both ◦ Auto ML ◦ Interpretable ML ◦ Machine discovery ◦ High-quality data quest
  29. Thank you tania.allard@microsoft.com