Coursera - How Google does ML
Mathematical Models used in ML
-
Neural Network
-
Linear methods
-
Decision trees
-
Radial basis functions
-
Ensembles of trees
-
Radial basis functions followed by linear methods
-
Support vector machines
-
Bagged decision trees
Learning Objectives
-
Build a data strategy around ML.
-
Discover where bias in machine learning models originates.
-
Explore Compute Engine and the basics of Cloud Storage.
-
Execute ad-hoc queries at scale.
-
Invoke pre-trained ML models from Datalab.Phases of modelling
Training phase
Prediction / Inference Phase
Applications
Topics
-
Training and serving skew
-
Inclusive ML
-
Equality of opportunity
-
How to find errors in your dataset using Facets
-
-
Bias
-
interaction bias (when a set of people are used they tend to draw generic things and computer then is biased towards non-generic things)
-
latent bias (when fed pictures of scientists, model can be biased towards men)
-
selection bias (select photos to train model from every place and not your album)
-
-
Ingest-transform-publish data
Path to ML
-
Individual Contributor
-
Delegation - Gently ramp up to include more people
-
Digitization - Automate mundane parts of the process
-
Big Data and Analytics - Measure and achieve data-driven success
-
Machine Learning - Automated feedback loop that can outpace human scale