Bayesian statistics, causal inference, and the web

I'm a data scientist at Google, where I work on time-series analysis of web traffic, anomaly detection, causal inference in large-scale experiments, and auction theory. Before joining Google, I was a postdoctoral researcher at the Department of Information Technology and Electrical Engineering at ETH Zurich working on Bayesian inference in high-dimensional time-series data.

One of the best things about statistical data analysis is how much progress we've seen in recent years in working with large time-series datasets. For example, there have been huge advances in approximate inference and increasingly powerful tools for running massively parallelized analyses. Some ideas, for example variational Bayesian techniques, have been popularized by computationally intensive disciplines such as computational neuroscience. The same concepts are really powerful to solve hard problems on web traffic and online advertising data.

Recent publications

  • Inferring causal impact using Bayesian structural time-series models
    K.H. Brodersen, F. Gallusser, J. Koehler, N. Remy, S. Scott (2015)
    Annals of Applied Statistics
  • CausalImpact, a new R package for estimating causal effects in time series
    Kay H. Brodersen / Google, Inc. (2014)
    Announcement   Project site   GitHub repository   Documentation  
  • Variational Bayesian mixed-effects inference for classification studies
    K.H. Brodersen, J. Daunizeau, C. Mathys, J.R. Chumbley, J.M. Buhmann, K.E. Stephan (2013)

Show all publications