Algorithmic and Statistical Perspectives on Large-Scale Data Analysis, Michael Mahoney

Описание: On Friday, October 10, 2014, Michael Mahoney, a senior researcher at ICSI and an associate adjunct professor in UC Berkeley's Statistics Department, spoke about large-scale data analysis. This talk was part of ICSI's annual research review. Read the full abstracts for this and other talks given at the review at https://www.icsi.berkeley.edu/icsi/ev...

Abstract:
Computer scientists have historically adopted quite different views on data (and thus on data management and data analysis) than statisticians, natural scientists, social scientists, and nearly everyone else who uses computation as a tool toward some downstream goal. For example, the former tend to view the data as noiseless bits and focus on algorithms with bounds on worst-case running time, independent of the input; while the latter typically have, either explicitly or implicitly, an underlying statistical model in mind and are interested in using computation and data to gain insight into the world. These issues are relevant now that “large-scale data analysis" has gone from being a technical topic of interest to a subset of computer scientists, to a cultural phenomenon that has a direct effect on nearly everyone. In this talk, I'll share some of my thoughts on these topics, I'll describe two applications (one in social network analysis and one in human genetics) where challenges related to these issues arose and describe how we dealt with them, and I'll offer some thoughts on how this so-called “Big Data" area might evolve.