SDS 451: Translating PhD Research into ML Applications — with Dan Shiebler

Ad Tech Models

Dan Shiebler discusses the intricate machine learning models used for advertising at Twitter, focusing on performance and revenue insights. He explains that performance ads are designed to prompt immediate actions, such as clicks or purchases, and involve complex machine learning processes to determine ad placement and value 1. Dan highlights the importance of software engineering in deploying and maintaining these models, noting that most work involves software engineering rather than model training 2.

The software engineering part is much more of it, especially at a company like Twitter, where the kind of size of our data, the speed with which our models need to respond to changes in the data distribution, requires enormous software lifts.

--- Dan Shiebler

This approach contrasts with smaller companies, where machine learning challenges are more prominent than software deployment issues.

   

No Code Tools

The rise of low-code and no-code tools is transforming the machine learning landscape, making it more accessible and efficient. Dan Shiebler shares his experience with BigQuery, a Google Cloud platform tool that simplifies machine learning model training with minimal configuration changes 3. He believes that these tools will not replace sophisticated infrastructure but will enable more people to focus on business problems rather than technical details 4.

This sort of low effort or low code or no code approaches to building new models that then go into all of these different places, that seems to me to be the future.

--- Dan Shiebler

This shift allows data professionals to allocate more time to innovative applications and strategic business initiatives.