Thu Jun 08 2023

Causal-driven ML & Cost-effective Transformer Productionization (Uber & Elsevier)

Raamstraat 7 · Amsterdam, NH


Join us for an evening of enlightening discussions and networking at our upcoming Data & Drinks on Thursday June 8th at 17:30 at Xomnia's HQ. We will explore two exciting topics reshaping the machine learning landscape: Productionizing transformer models with minimal cost using open-source ecosystems at Elsevier, and Causal-driven machine learning at Uber.

The event includes dinner, drinks and a lot of networking opportunities with data professionals from Amsterdam and beyond.

Summary of the talks:

Talk #1: Productionize transformer models with minimal cost using open-source ecosystems
Presently, 90% of transformer models are in the PoC stage. The performance of transformer models is great but they are not pushed to production due to higher operation cost, maintenance problems, and complex integration problems. Elsevier solved the problem with KServe and Kubeflow pipelines, and achieved under 1-millisecond latency of the Bert (not Distillbert) model with minimal cost. Elsevier's MLOPs Lead Rahul Dutta will dive into how this was achieved in his talk.

Talk #2: Causal-driven Machine Learning at Uber
During this session, Uber's Ed Lo and Okke van de Wal will share their invaluable insights on how they introduced causality into Uber's machine learning models, enabling the creation of highly personalized user experiences. Through captivating case studies, they will illustrate the successful combination of experimental data and machine learning to achieve unprecedented levels of personalization.

Furthermore, they will shed light on CausalML, an open-source Python package developed at Uber. Our audience will gain an understanding of how this powerful tool can aid in the seamless transition from correlation-driven machine learning to causal-driven machine learning. Don't miss this opportunity to learn from their expertise and unlock the potential of causal-driven ML for your own projects.

About the speakers:

Raahul Dutta: Raahul is MLOPs Lead at Elsevier, with over 6 years of experience in transforming the Jupyter Notebooks into low-latency, highly scalable, production-standard endpoints. He implemented various ML/AI models and pipelines (30+) and exposed them. He was also associated with Oracle, UHG, and Philips, and filed around 13 patents (some of them in `granted` status) in the ML, BMI and chatbot domain. Raahul enjoys the thrill of riding motorbikes and resides in Amsterdam with his partner.

Ed Lo and Okke van de Wal: Ed Lo (Manager, Product Data Science at Uber) and Okke van der Wal (Manager, Machine Learning at Uber) specifically focus on payments, leveraging techniques such as machine learning, causal inference & experimentation projects. Projects they have worked on include Anomaly Detection, Personalization & Fraud Detection within the Onboarding and Checkout experiences in the Uber apps.