Python Machine Learning Conference & GeoPython 2020

»Democratizing Machine Learning Model Development«
2020-09-22, 09:15–09:45, Room 2

Despite the availability of various modern deployment infrastructures - serverless, containers, servers, on-device; developers are often constrained by specific framework or programming language as governed by their firm's deployment infrastructure. This talk will focus on the emergence of various Model Exchange Formats which are democratizing the process of model development.

Data is the new Oil! But, what is its use if you cannot refine (analyse) & sell (derive value) it. Big Data has pushed the frontier of analytical processing from having analytical servers to performing analytics close to Data source.

Model development and Model deployment are the two fundamental components of an analytics life-cycle. Today, various deployment options are available like serverless, containers, servers and on-device, but developers are constrained by specific framework or programming language as governed by their firm's deployment infrastructure. This talk will focus on the emergence of various Model Exchange Formats which are democratizing the process of model development by allowing users to choose the suitable framework for their model development process without the pangs of deployment infrastructure.

The talk will cover the following sections:

  • Evolution of analytics (Dedicated Machines -> Cloud -> Edge) - 3 mins
  • The need of Edge analytics and use cases - 3 mins
  • Analytics Life-cycle (ALC): Introduction, Importance of Model Deployment - 4 mins
  • Model Deployment Architectures - 5 mins
  • Model Exchange Formats (PMML, PFA, ONNX) for Deployment: Introduction & Need for Democratizing model development process - 5 mins
  • Example Model Interchange Format - Portable Format for Analytics (PFA) - 5 mins
  • Q&A - 5 mins