Abstract: 

Historically, data scientists looking to do cutting-edge machine learning have had to choose between implementing their entire models and pipelines from scratch using lower-level APIs such as Tensorflow and PyTorch or using simpler but much more limited black-box AutoML technologies. In this session, we introduce an alternative to these two choices which provides the best of flexibility and simplicity through a configuration-based modeling interface referred to as declarative machine learning. We discuss how declarative machine learning interfaces have been at the heart of fast ML adoption at leading technology companies like Meta (Facebook), Uber and Apple and provide a deep dive into Ludwig – an open source declarative machine learning framework that allows citizen data scientists to specify entire cutting-edge machine learning pipelines in just a few lines of code.

During the talk, we’ll show how Ludwig’s novel compositional model architecture referred to as encoder-combiner-decoder makes it possible to easily mix multiple modalities of data such as text, images, audio with structured data in a way that is consistently easy across tasks like regressions, classification, and even generation. We’ll also show how the configuration-driven interface makes building state-of-the-art deep learning models accessible to data scientists of all backgrounds, by providing fine-tuning capabilities for best-in-class transformer architectures like BERT, T5, VIT and many more. Finally, we’ll end with how declarative machine learning can also simplify some of the infrastructure requirements around training and operationalizing machine learning models and what’s next on our open source roadmap to make this easier for our community to engage with.

Bio: 

Dev is co-founder and Chief Product Officer for Predibase, a company looking to redefine how data scientists and engineers build models with a declarative approach. Prior to Predibase, he was a ML PM at Google working across products like Firebase, Google Research and the Google Assistant as well as Vertex AI. While there, Dev was also the first product manager for Kaggle – a data science and machine learning community with over 8 million users worldwide. Dev’s academic background is in computer science and statistics, and he holds a masters in computer science from Harvard University focused on machine learning.