The Machine Learning Factory: Automation of the ML Lifecycle
Speaker:
Jason Barto, AWS Solutions Architect, AWS
The lifecycle of a machine learning model, and more importantly the business insights it offers, is an iterative and ever evolving process. From feature discovery and engineering, to model training and selection, even through to production hosting and drift detection, AWS services can support and automate the events that lead to change in a customer’s model.
Join us to see a demonstration of how AWS services can be used to transform raw data into an engineered feature set that then triggers the training and evaluation of an updated model. This session will address topics such as context drift, secure hosting of trained models as a RESTful API, and automation for retraining models when data or code changes.
3. “The most radical and transformative
of inventions are those
that empower others
to unleash their creativity
- to pursue their dreams.”
– Jeff Bezos
4. AWS Pace of Innovation
AWS has been continually expanding its services to
support virtually any cloud workload, and it now has
more than 90 services that range from compute,
storage, networking, database, analytics, application
services, deployment, management, developer, mobile,
Internet of Things (IoT), Artificial Intelligence (AI),
security, hybrid and enterprise applications. AWS has
launched a total of 1,430 new features and/or services
year to date* for a total of 4,343 new features and/or
services since inception in 2006.
2011
82
722
1,430
280
2013 2015 2017
* As of 1 January 2018
6. Machine learning process is hard
Fetch data
Clean &
format data
Prepare &
transform
data
Train model
Evaluate
model
Integrate
with prod
Monitor /
debug /
refresh
Data wrangling
• Set up and
manage data
workbench
• Get data to
workbench
securely
Experimentation
• Set up and
manage clusters
• Scale / distribute
ML algorithms
Deployment
• Set up and
manage inference
clusters
• Manage and auto
scale inference
APIs
• Testing,
versioning, and
monitoring
16. Machine learning on AWS
ü Security
ü Feature engineering
ü Tool agnostic
ü Project traceability and auditability
ü Hyperparameter auto-tuning
ü Ensemble of models
ü Context drift
17. AI and ML at Intuit have three areas of focus.
ü Smart products
ü Fraud detection and prevention
ü Customer care and expert advice
Real-time Fraud Detection with Amazon SageMaker
In order to keep fraudsters out of their systems and data,
Intuit always stays several moves ahead by leveraging
AI/ML-generated insights from data that can determine
real-time fraud detection in TurboTax: Specifically
• Account take-over detection at login
• Identity theft detection at filing