Transdisciplinary Foundations of Data Science

The goal of this workshop is to explore transdisciplinary foundations of data science and to explore the creation of a research community to continue discussion beyond the workshop.

Background

Despite rapid growth in data science applications, there is inadequate progress on data science foundations. Lack of a strong foundation and understanding of scientific issues (e.g., generalizability, reproducibility, computability, and prediction error bounds) makes it difficult for practitioners to develop reliable models from data and for the public to trust the models derived from big data. For example, despite access to tremendous volumes of data and computational resources as well as early successes, the Google flu trend analysis led to major disappointments, which are described in the recent articles in Nature, Science and PLOS. In addition, data science applications are facing public criticism related to privacy, discrimination, and equity.

Workshop Activities 

This workshop will bring together researchers investigating data science foundations across multiple disciplines, including mathematics, statistics, and computer science. The activities will start with surveying discipline-specific perspectives, results, and knowledge gaps. This will be followed by panel discussions to explore a transdisciplinary, holistic approach integrating and moving beyond discipline-specific approaches. The workshop will explore modes of collaboration to continue the transdisciplinary approach to data science foundations beyond this workshop. A possible outcome is a white paper.

Topics include the following:

• Scientific methods (e.g., reproducibility of data science results)
• Mathematical foundations (e.g., performance guarantees, probabilistic estimates for random models, generalization bounds)
• Statistical foundations (e.g., causal inference, uncertainty quantification, post model selection inference)
• Computing foundations (e.g., computational complexity of data science problems and methods, scalable algorithms, resource trade-offs) 
• Privacy-aware data science (e.g., differential privacy)  
• Potential negative societal impacts of data science methods, such as discrimination, disparity, and racial bias
• Transdisciplinary foundations