Florian Gerber

Florian Gerber


Applied Mathematics and Statistics Department, Colorado School of Mines, 1500 Illinois St., Golden, CO 80401, USA
gerber@mines.edu

Applied Mathematics and Statistics Department
Colorado School of Mines
1500 Illinois St., Golden, CO 80401, USA
gerber@mines.edu



I develop cutting-edge machine learning methods and software for the analysis of large datasets.


Curriculum vitae


Experience

2018— now: PostDoc at the Applied Mathematics and Statistics Department, Colorado School of Mines, Golden CO, USA
Visiting scientist at the National Center for Atmospheric Research, Boulder CO, USA
2013—2018: PhD assistant and PostDoc Department of Mathematics, University of Zurich, Switzerland
2012—2013: Junior statistician at the Institute of Social and Preventive Medicine, University of Bern, Switzerland
2011—2013: External statistics and R programming consultant for Novartis Pharma AG, Basel, Switzerland

Education

2017: PhD in Applied Statistics, University of Zurich, Switzerland
2013: MSc in Biostatistics, University of Zurich, Switzerland
2010: BSc in Mathematics with minor in Philosophy, University of Bern, Switzerland
2009: Certificate in Interdisciplinary Studies in Ecology, University of Bern, Switzerland
2006: Matura, Gymnasiums Biel-Seeland, Switzerland

More information is given in the CV and on LinkedIn .
More information is given in the CV and on LinkedIn .


Research

  • develop cutting-edge machine learning methods for large datasets
  • high-performance, cloud computing, statistical software
  • Gaussian processes, Bayesian inference, neural networks, deep learning
  • Image processing for remote sensing, climate model, and medical applications

I. Use neural networks for fast covariance parameter estimation

  • reduces the computing time for covariance parameter estimation of Gaussian process models by at least a factor of 100
  • the TensorFlow implementation features a customized data generator defined using the Python API
  • data illustration using climate model output from the LENS project
Gerber F., Nychka D. W. Fast covariance parameter estimation of spatial Gaussian process models, 2021
scaling

II. Parallel computing for Gaussian process models

  • a Gaussian process model is fitted to a spatial dataset with 5 million observations
  • high-performance computing with 512 CPUs in parallel is used
  • uses cross-validation and the division of the domain in overlapping subsets
Gerber F., Nychka D. W. Parallel cross-validation: a scalable fitting method for Gaussian process models, Computational Statistics & Data Analysis, 2020

scaling

III. Predicting missing values in spatio-temporal remote sensing data

  • predictions are obtained by selecting subsets, sorting procedures, and quantile regression
  • the workload can be distributed to several computers
  • the method is available in the R package gapfill available on CRAN
Gerber F., de Jong R., Schaepman M. E., Schaepman-Strub G., Furrer R. Predicting Missing Values in Spatio-Temporal Remote Sensing Data, IEEE Transactions on Geoscience and Remote Sensing, 2018
gapfill

IV. 64-bit sparse matrices with the R package spam

  • the R packages spam and spam64 provide methods for sparse matrices with more than 2^31 non-zero elements
  • the R package dotCall64 provides an enhanced interface to link R with compiled code
  • the new functionality of the R packages is illustrated with a non-stationary spatial model
Gerber F., Mösinger K., Furrer R. Extending R Packages to Support 64-bit Compiled Code: An Illustration with spam64 and GIMMS NDVI3g Data, Computers & Geoscience, 2016

Gerber F., Mösinger K, Furrer R. dotCall64: An Efficient Interface to Compiled C/C++ and Fortran Code Supporting Long Vectors, SoftwareX, 2018

spam64

V. Bayesian inference and medical applications

  • assessment of inference methods for Bayesian hierarchical models
  • spatial models for areal count data
  • R package gsbDesign for operating characteristics of Bayesian clinical trial designs
Gerber F., Furrer R. Pitfalls in the Implementation of Bayesian Hierarchical Modeling of Areal Count Data: An Illustration Using BYM and Leroux Models, Journal of Statistical Software, 2015
Gerber F., Gsponer T. gsbDesign: An R Package for Evaluating the Operating Characteristics of a Group Sequential Bayesian Design, Journal of Statistical Software, 2016
Gsponer T., Gerber F., Bornkamp B., Ohlssen D., Vandemeulebroecke M., Schmidli H. Bayesian Group Sequential Designs for Clinical Trials and their Operating Characteristics, Pharmaceutical Statistics, 2014

mcmc



Software


R packages on CRAN

optimParallel: Parallel versions of the L-BFGS-B optimizer
gapfill: Fill missing values in satellite data
gsbDesign: Operating characteristics for clinical, Bayesian trials designs
spam: Sparse matrix algebra
spam64: 64-bit extension of spam
dotCall64: Enhanced foreign function interface supporting long vectors
fields: Tools for spatial data
vcd: Tools for categorical data

Python package on PyPI

optimparallel: Parallel versions of L-BFGS-B optimization method in scipy

Shiny R web application

Shut the Box: A variant of the dice game

R packages on CRAN

optimParallel Parallel versions of the L-BFGS-B optimizer
gapfill Fill missing values in satellite data
gsbDesign Operating characteristics for clinical, Bayesian trials designs
spam Sparse matrix algebra
spam64 64-bit extension of spam
dotCall64 Enhanced foreign function interface supporting long vectors
fields Tools for spatial data
vcd Tools for categorical data

Python package on PyPI

optimparallel Parallel versions of L-BFGS-B optimization method in scipy

Shiny R web application

Shut the Box A variant of the dice game


Publications


[16] Gerber, F. and Nychka, D. W. Fast covariance parameter estimation of spatial Gaussian process models. Under review, 2021. arXiv:2012.15339

[15] Roman, F. Gerber, F., Schmid, B, and Furrer, R. Identification of dominant features in spatial data. Spatial Statistics, 2020 doi: 10.1016/j.spasta.2020.100483

[14] Gerber, F., and Nychka, D. W. Parallel cross-validation: a scalable fitting method for Gaussian process models. Computational Statistics & Data Analysis, 2020. doi: 10.1016/j.csda.2020.107113

[13] Gerber, F., and Furrer, R. optimParallel: An R package providing a parallel version of the L-BFGS-B optimization method. The R Journal, 11(1):352–358, 2019. doi: 10.32614/RJ-2019-030

[12] Heaton, M. J., Datta, A., Finley, A., Furrer, R., Guinness, J., Guhaniyogi, R., Gerber, F., Gramacy, R. B., Hammerling, D., Katzfuss, M., Lindgren, F., Nychka, D. W., Sun, F., and Zammit-Mangion, A. A case study competition among methods for analyzing large spatial data. Journal of Agricultural, Biological and Environmental Statistics, 1–28, 2018. doi: 10.1007/s13253-018-00348-w

[11] Warren, B. H., Hagen, O., Gerber, F., Thébaud, C., Paradis, E., and Conti, E. Evaluating alternative explanations for an association of extinction risk and evolutionary uniqueness in multiple insular lineages. Evolution, 72:2005–2024, 2018. doi: 10.1111/evo.13582

[10] Gerber, F., Mösinger, K., and Furrer, R. dotCall64: An efficient interface to compiled C/C++ and Fortran code supporting long vectors. SoftwareX, 7:217–221, 2018. doi: 10.1016/j.softx.2018.06.002

[9] Gerber, F., de Jong, R., Schaepman, M. E., Schaepman-Strub, G., and Furrer, R. Predicting missing values in spatio-temporal remote sensing data. IEEE Transactions on Geoscience and Remote Sensing, 56(5):2841–2853, 2018. doi: 10.1109/TGRS.2017.2785240

[8] Gerber, F., Mösinger, K., and Furrer, R. Extending R packages to support 64-bit compiled code: An illustration with spam64 and GIMMS NDVI3g data. Computers & Geosciences, 104:109–119, 2017. doi: 10.1016/j.cageo.2016.11.015

[7] Gerber, F. and Gsponer, T. gsbDesign: An R package for evaluating the operating characteristics of a group sequential Bayesian design. Journal of Statistical Software, 69(1):1–23, 2016. doi: 10.18637/jss.v069.i11

[6] Bratulic, S., Gerber, F., and Wagner, A. Mistranslation drives the evolution of robustness in tem-1 β-lactamase. Proceedings of the National Academy of Sciences, 112(41):12758–12763, 2015. doi: 10.1073/pnas.1510071112

[5] Gerber, F. and Furrer, R. Pitfalls in the implementation of Bayesian hierarchical modeling of areal count data: An illustration using BYM and Leroux models. Journal of Statistical Software, 63(1):1–32, 2015. doi: 10.18637/jss.v063.c01

[4] Bashir, T., Sailer, C., Gerber, F., Loganathan, N., Bhoopalan, H., Eichenberger, C., Grossniklaus, U., and Ramamurthy, B. Hybridization alters spontaneous mutation rates in a parent-of-origin-dependent fashion in Arabidopsis. Plant Physiology, 165(1):424–437, 2014. doi: 10.1104/pp.114.238451

[3] Gsponer, T., Gerber, F., Bornkamp, B., Ohlssen, D., Vandemeulebroecke, M., and Schmidli, H. A practical guide to Bayesian group sequential designs. Pharmaceutical Statistics, 13(1):71–80, 2014. doi: 10.1002/pst.1593

[2] Gerber, F., Marty, F., Eijkel, G. B., Basler, K., Brunner, E., Furrer, R., and Heeren, R. M. A. Multiorder correction algorithms to remove image distortions from mass spectrometry imaging data sets. Analytical Chemistry, 85(21):10249–10254, 2013. doi: 10.1021/ac402018e

[1] Wandeler, G., Gerber, F., Rohr, J., Chi, B., Orrell, C., Chimbetete, C., Prozesky, H., Boulle, A., Hoffmann, C., Gsponer, T., Fox, M., Zwahlen, M., Egger, M., and IeDEA S. A. Tenofovir or Zidovudine in second-line antiretroviral therapy after Stavudine failure in Southern Africa. Antiviral Therapy, 19(5):521–525, 2013. doi: 10.3851/imp2710