TF . Probability

for HyperParameters


TensorFlow & Deep Learning SG


Martin Andrews @ reddragon.ai

10 October 2019

About Me

  • Machine Intelligence / Startups / Finance
    • Moved from NYC to Singapore in Sep-2013
  • 2014 = 'fun' :
    • Machine Learning, Deep Learning, NLP
    • Robots, drones
  • Since 2015 = 'serious' :: NLP + deep learning
    • & GDE ML; TF&DL co-organiser
    • & Papers...
    • & Dev Course...

About Red Dragon AI

  • Google Partner : Deep Learning Consulting & Prototyping
  • SGInnovate/Govt : Education / Training
  • Products :
    • Conversational Computing
    • Natural Voice Generation - multiple languages
    • Knowledgebase interaction & reasoning

Outline

  • whoami = DONE
  • TensorFlow Probability
  • Mathematics
  • Notebook 1 : Model fitting
  • Notebook 2 : Hyperparameter Search
  • Wrap-up

TF . probability

  • Open source library on top of TensorFlow
  • Knows all about Probabilities and Distributions
    • ... and build models from them
  • Modelling "the right way"
    • (i.e. mathematician approved)

Mathematics 1

  • A line has the equation :
  • $$y=m.x+c$$
  • Or (for neural networks) :
  • $$y=w.x+b$$
y=mx+c

Mathematics 2

  • A normal distribution looks like :
  • $$\mathcal{N}(\mu, \sigma^2) :: P(x) \sim e^{-\frac{(x-\mu)^2}{2\sigma^2}}$$
Normal Distribution (from Wikipedia)

Probabilistic Modelling

  • Fitting some points with model(s)
  • Notebook 1

Hyperparameter Search

  • More Gaussian Processes
  • ( and a CNN for CIFAR10 )
  • Notebook 2

Wrap-up

  • Only a glimpse of what can be done
  • Warning : Mathematics!
  • Wider scope than "just" Deep Learning
GitHub - mdda

* Please add a star... *

Deep Learning
MeetUp Group

Deep Learning : Jump-Start Workshop

Deep Learning
Developer Course

- QUESTIONS -


Martin @
RedDragon . AI