Abstract: In recent years, we have seen amazing results in artificial intelligence and machine learning owing to the emergence of models such as transformers and pretrained language models. Despite the astounding results published in academic papers, there remains a lot of ambiguity and challenges when it comes to deploying these models in industry because: 1) troubleshooting, training, and maintaining these models is very time and cost consuming due to their inherent large sizes and complexities 2) there is not yet enough clarity about when the advantages and challenges of these models outweigh classical ML models. These challenges are even more severe for small and mid-sized companies that do not have access to huge compute resources and infrastructure. In this talk, we discuss these challenges and share our findings and recommendations from working on real world examples at SPINS, a data/tech company focused on the natural grocery industry. More specifically, we describe how we leverage state-of-the-art language models to seamlessly automate parts of SPINS’ data ingestion workflow and drive substantial business outcomes. We provide a walk through of our end-to-end MLOps system and discuss how using the right tools and methods have helped to mitigate some of these challenges. We also share our findings from our experimentation and provide insights on when one should use these massive transformer models instead of classical ML models. Considering that we have a variety of challenges in our use cases from an ill-defined label space to a huge number of classes (~86,000) and massive data imbalance, we believe our findings and recommendations can be applied to most real-world settings. We hope that the learnings from this talk can help you to solve your own problems more effectively and efficiently!
Bio: Azin Asgarian is currently an applied research scientist on Georgian’s R&D team where she works with companies to help adopt applied research techniques to overcome business challenges. Prior to joining Georgian, Azin was a research assistant at the University of Toronto and part of the Computer Vision Group where she was working on the intersection of Machine Learning, Transfer Learning, and Computer Vision.