The variational approximation of posterior distributions by multivariate Gaussians has been much less popular in the Machine Learning community compared to the corresponding approximation by factorising distributions. This is for a good reason: the Gaussian approximation is in general plagued by an O(N 2 ) number of variational parameters to be optimised, N being the number of random variables. In this work, we discuss the relationship between the Laplace and the variational approximation and we show that for models with Gaussian priors and factorising likelihoods, the number of variational parameters is actually O(N ). The approach is applied to Gaussian process regression with non-Gaussian likelihoods.

Abstract. Side-channel attacks are a serious threat to implementations of cryptographic algorithms. Secret information is recovered based on power consumption, electromagnetic emanations or any other form of physical information leakage. Template attacks are probabilistic sidechannel attacks, which assume a Gaussian noise model. Using the maximum likelihood principle enables us to reveal (part of) the secret for each set of recordings (i.e., leakage trace). In practice, however, the major concerns are (i) how to select the points of interest of the traces, (ii) how to choose the minimal distance between these points, and (iii) how many points of interest are needed for attacking. So far, only heuristics were provided. In this work, we propose to perform template attacks in the principal subspace of the traces. This new type of attack addresses all practical issues in principled way and automatically. The approach is validated by attacking stream ciphers such as RC4. We also report analysis results of template style attacks against an FPGA implementation of AES Rijndael. Roughly, the template attack we carried out requires five time less encrypted messages than the best reported correlation attack against similar block cipher implementations.

Background: Alanine scanning mutagenesis is a powerful experimental methodology for investigating the structural and energetic characteristics of protein complexes. Individual aminoacids are systematically mutated to alanine and changes in free energy of binding (ΔΔG) measured. Several experiments have shown that protein-protein interactions are critically dependent on just a few residues ("hot spots") at the interface. Hot spots make a dominant contribution to the free energy of binding and if mutated they can disrupt the interaction. As mutagenesis studies require significant experimental efforts, there is a need for accurate and reliable computational methods. Such methods would also add to our understanding of the determinants of affinity and specificity in protein-protein recognition.

Abstract. The power consumption and electromagnetic radiation are among the most extensively used side-channels for analyzing physically observable cryptographic devices. This paper tackles three important questions in this respect. First, we compare the effectiveness of these two side-channels. We investigate the common belief that electromagnetic leakages lead to more powerful attacks than their power consumption counterpart. Second we study the best combination of the power and electromagnetic leakages. A quantified analysis based on sound information theoretic and security metrics is provided for these purposes. Third, we evaluate the effectiveness of two data dimensionality reduction techniques for constructing subspace-based template attacks. Selecting automatically the meaningful time samples in side-channel leakage traces is an important problem in the application of template attacks and it usually relies on heuristics. We show how classical statistical tools such as Principal Component Analysis and Fisher Linear Discriminant Analysis can be used for efficiently preprocessing the leakage traces.

A new variational Bayesian learning algorithm for Student-t mixture models is introduced. This algorithm leads to (i) robust density estimation, (ii) robust clustering and (iii) robust automatic model selection. Gaussian mixture models are learning machines which are based on a divide-and-conquer approach. They are commonly used for density estimation and clustering tasks, but are sensitive to outliers. The Student-t distribution has heavier tails than the Gaussian distribution and is therefore less sensitive to any departure of the empirical distribution from Gaussianity. As a consequence, the Student-t distribution is suitable for constructing robust mixture models. In this work, we formalize the Bayesian Student-t mixture model as a latent variable model in a different way than Svensén and Bishop (2004). The main difference resides in the fact that it is not necessary to assume a factorized approximation of the posterior distribution on the latent indicator variables and the latent scale variables in order to obtain a tractable solution. Not neglecting the correlations between these unobserved random variables leads to a Bayesian model having an increased robustness. Furthermore, it is expected that the lower bound on the log-evidence is tighter. Based on this bound, the model complexity, i.e. the number of components in the mixture, can be inferred with a higher confidence.