As machine learning (ML) becomes the centerpiece of innumerable applications with direct impact on human beings, being able to explore, understand, test and debug models has become a crucial aspect for machine learning adoption. Especially when ML is employed for critical applications that impact people's lives (e.g., medical interventions and hiring decisions) , it is essential for data scientists and domain experts to have a better sense of how a model may behave once it is deployed.

In this talk, I describe the work we are doing to support exploratory data analysis of machine learning data and models through the development of interactive data visualization. I will explain why exploratory analysis of models is necessary and what we can achieve with it. I will also describe techniques and tools we developed to support the work of data scientists and domain experts in this area. I will conclude with a discussion of open issues and hints of technologies we will need in the near future.


bio:Enrico Bertini is an Associate Professor at Northeastern University with a double appointment between Computer Science (Khoury) and Art&Design (CAMD). Before joining Northeastern he was at New York University, where has been a professor between 2012-2021. His area of research is data visualization, with a focus on data visualization for machine learning, computational guidance for data exploration and the cognitive/perceptual aspects of data visualization. Prof. Bertini is also the co-host of Data Stories, a popular podcast on the role data play in our lives.