The week of June 20th was a busy time for machine learning in New York City; both the ICML 2016 (which more than doubled in size compared to ICML 2015) and UAI 2016 conferences were held in the city, and Bloomberg’s machine learning teams participated in both events. We also hosted a reception at our headquarters for ICML attendees, and Andrew Ng (of Stanford, Baidu, and Coursera) spoke to members of our engineering teams. As you can see, machine learning is a very active research and development area at Bloomberg, and we’re proud to participate in the global community.
Monday, June 20 - Bloomberg-ICML Reception
Over 400 people from 200+ organizations enjoyed hours of interaction and networking with their peers in the machine learning community.
Wednesday, June 22 - Andrew Ng speaks to Bloomberg's machine learning teams
Friday, June 24 – #Data4Good Workshop at ICML 2016
Gideon Mann and Mark Dredze of Bloomberg’s machine learning team presented their paper “Twitter as a Source of Global Mobility Patterns for Social Good” at the Data4Good workshop at ICML 2016.
Data on human spatial distribution and movement is essential for understanding and analyzing social systems. Yet existing sources for these data are lacking in various ways; difficult to access, biased, have poor geographical or temporal resolution, or are significantly delayed. New sources of global mobility patterns that are easy and fast to access, cover both local and global areas, and can scale are needed for a variety of applications, including epidemiological models of disease spread and responding to human migration due to economic hardship and political unrest.
As part of our news and social media products, we have a unique perspective on Twitter data. Bloomberg has been using and analyzing Twitter data for some time now, enabling our clients to better understand global sentiment and market reactions. Building on our expertise, we are using geocoded Twitter data to produce an estimate of global travel patterns.
Together with data scientists at UNICEF, we are studying how Twitter data can estimate global mobility patterns and address shortcomings of existing methods. These findings will inform how this novel data source can be harnessed to address humanitarian and development efforts. This paper is our first step at addressing data needs that can have a meaningful impact on public health and safety.
Saturday, June 25 – Conference on Uncertainty in Artificial Intelligence
Dr. Anshul Gupta of the IBM TJ Watson Research Center and Dr. Prabhanjan Kambadur of Bloomberg presented their paper “Parallel and High-performance Computing for Speeding up Machine Learning Algorithms” at UAI 2016.
Their talk gave a generic introduction to parallel and high-performance computing, including hardware, parallel programming models, parallel programming tools, and most common challenges and pitfalls. They also discussed performance and scalability metrics, parallel algorithm analysis, and the importance of such analysis. The paper illustrates some of the parallelization and analysis techniques with a case study involving an end-to-end machine learning classification problem.