Anju Kambadur, Ania Musial, Ian Hummel
Bloomberg L.P.
Invited Talk: The Bloomberg Data Science Platform
Abstract:
The Bloomberg Professional Service (aka “the Terminal”) provides data, analytics, news, information and communication to professionals in business, finance, government, and philanthropy. Over the past decade, Bloomberg has invested heavily in three specialized areas in the field of AI: natural language processing (or, the application of machine learning methods to text), information retrieval and search, and core machine learning, including deep learning.
In order to facilitate the model development life cycle, we have built an internal Data Science Platform (DSP). Now, our AI research scientists can run many large experiments quickly: it’s easy to define machine learning jobs and the dataset they want to analyze, train their models and evaluate the results, and then fine-tune their models to improve the outcomes, without worrying about what hardware or software needs to be managed. Once models are trained, they can deploy an inference service in seconds with one click using KFServing. All of the best tools – from GPUs and Spark/Kubernetes clusters for compute to TensorFlow and PyTorch libraries – are there for them to use.
In this talk, we’ll introduce the DSP, how we evaluated and designed its core components, and finally showcase how our DSP has delivered massive improvements in efficiency, enabling our teams to shorten their model training on existing data from months to hours.
Biography - Anju Kambadur:
Anju Kambadur is the head of the AI Engineering at Bloomberg, which consists of 150+ researchers and engineers building solutions for Finance using technologies from Machine Learning, Natural Language Processing, Information Retrieval, Recommendation Systems, Vision, and Optimization. Some of the products the AI Group helps build include Bloomberg News, Research, Communications, and Finance. Before Bloomberg, Anju was a research staff member at IBM Research’s Thomas J. Watson Research Center, where he worked on problems in machine learning, such as matrix sketching, genome-wide association studies, temporal causal modeling, and high performance computing. He received his Ph.D from Indiana University. Anju has published peer-reviewed articles in the fields of high performance computing, machine learning, and natural language processing.
Biography - Ania Musial:
Ania Musial is a Senior Software Engineer in the AI Engineering group at Bloomberg, focusing on components such as experiment management, hyperparameter tuning, and distributed training that are required for a robust machine learning model development lifecycle. Previous projects in her 11 year tenure at Bloomberg include building recommendation system infrastructure, development of real-time news applications, and advocacy for engineers’ needs. She is the chair of Bloomberg’s Machine Learning Guild, which is responsible for creating a unified ML community and promoting discussions of ML problems, techniques, and best practices across Bloomberg. She has a BS in Mathematical Science and Linguistics from University of Michigan, and has given talks at KubeCon and Open FinTech Forum.
Biography - Ian Hummel:
Ian Hummel is the Product Manager of Bloomberg’s Data Science Platform. As part of his role in the Office of the CTO, he oversees strategic initiatives at the intersection of machine learning, big data, and cloud computing. Previous to Bloomberg, he led product initiatives in online marketing, identity federation, and enterprise search. He has a BA in Mathematics and Computer Science from Boston University and an MBA from INSEAD.