• Ogden, H. E. (2015). A sequential reduction method for inference in generalized linear mixed models. Electronic Journal of Statistics, 9(1), pp. 135 - 152.
  • The likelihood for the parameters of a generalized linear mixed model involves an integral which may be of very high dimension. This paper describes a method to exploit the structure of the integrand to reduce substantially the cost of finding an accurate approximation to the likelihood in many of these models.  Abstract / Download (PDF)

  • Ogden, H. E. (2015). A caveat on the robustness of composite likelihood estimators: the case of mis-specified random effect distribution. Statistica Sinica, to appear.
  • A  composite  likelihood  estimator  will  be  robust  to  certain types  of model misspecification, and this potential for increased robustness is considered a secondary motivation for the use of composite likelihood. The purpose of this paper is to show that there are some situations in which a composite likelihood estimator may actually suffer a loss of robustness compared to the maximum likelihood estimator.  Download (PDF)

  • Aslett, L. J. M., Esperança, P. M. and Holmes, C. C. (2015). A review of homomorphic encryption and software tools for encrypted statistical machine learning, Technical report, University of Oxford. arXiv:1508.06574 [stat.ML].
  • Recent advances in cryptography promise to enable secure statistical computation on encrypted data, whereby a limited set of operations can be carried out without the need to first decrypt. We review these homomorphic encryption schemes in a manner accessible to statisticians and machine learners, focusing on pertinent limitations inherent in the current state of the art.  Abstract / Download (PDF)

  • Aslett, L. J. M., Esperança, P. M. and Holmes, C. C. (2015). Encrypted statistical machine learning: new privacy preserving methods. arXiv:1508.06845 [stat.ML]
  • We present two new statistical machine learning methods designed to learn on fully homomorphic encrypted (FHE) data.  We demonstrate that these techniques perform competitively on a variety of classification data sets and provide detailed information about the computational practicalities of these and other FHE methods.  Abstract / Download (PDF)

  • Software: HomomorphicEncryption and EncryptedStats R packages
  • The first provides easy to use high level language access to homomorphic encryption via R.  All common operations are overloaded with fully native support for vectors and matrices, so that cipher texts can be manipulated exactly like ordinary data.  The second package provides implementations of novel machine learning algorithms which can run within the constraints of homomorphic encryption.

  • This work examines the asymptotic properties of approximate Bayesian computation methods and proposes an iterative algorithm for finding a sensible importance proposal distribution.  Abstract / Download (PDF)

  • The likelihood for the parameters of a generalized linear mixed model involves an integral which may be of very high dimension. This paper describes a method to exploit the structure of the integrand to reduce substantially the cost of finding an accurate approximation to the likelihood in many of these models.  Abstract / Download (PDF)

  • A  composite  likelihood  estimator  will  be  robust  to  certain types  of model misspecification, and this potential for increased robustness is considered a secondary motivation for the use of composite likelihood. The purpose of this paper is to show that there are some situations in which a composite likelihood estimator may actually suffer a loss of robustness compared to the maximum likelihood estimator.  Download (PDF)

  • Recent advances in cryptography promise to enable secure statistical computation on encrypted data, whereby a limited set of operations can be carried out without the need to first decrypt. We review these homomorphic encryption schemes in a manner accessible to statisticians and machine learners, focusing on pertinent limitations inherent in the current state of the art.  Abstract / Download (PDF)

  • We present two new statistical machine learning methods designed to learn on fully homomorphic encrypted (FHE) data.  We demonstrate that these techniques perform competitively on a variety of classification data sets and provide detailed information about the computational practicalities of these and other FHE methods.  Abstract / Download (PDF)

Software: HomomorphicEncryption and EncryptedStats R packages

  • The first provides easy to use high level language access to homomorphic encryption via R.  All common operations are overloaded with fully native support for vectors and matrices, so that cipher texts can be manipulated exactly like ordinary data.  The second package provides implementations of novel machine learning algorithms which can run within the constraints of homomorphic encryption.