Persyvact

Scientific description of the project

The goal of the Persyvact research project is to develop tools for analyzing hierarchically structured models of complex, high-dimensional data. The tools offered by Persyvact are based on cutting-edge mathematical and algorithmic developments. A universal idea underlying the concept of structure is that combining simple local components into a coherent model allows modelers to describe complex data with great precision.

The challenge of analyzing "big" datasets also requires researchers in applied sciences, biology, medicine, and signal processing to collaborate with researchers in computational mathematics to enhance the capabilities of their traditional methods and address new questions.

Several traditional calculation methods are in fact limited by overly restrictive assumptions, and their lack of robustness introduces errors that can skew data analyses.

This is the case when samples exhibit complex statistical dependencies, for example due to repeated experiments, uneven experimental designs, clustered or grouped data, and spatiotemporal relationships. A second source of complexity stems from the measurement process itself, which may involve very different instruments or record data of very different types. High-dimensional data are often heterogeneous and potentially noisy or missing, and they may contain multiscale information across the spatial and temporal scales of the observed process.

The Persyvact project proposes mathematical and statistical methods to facilitate the analysis of high-dimensional data characterized by complex dependencies and heterogeneous or multilevel structures. It addresses key issues in the analysis of large-scale numerical data, population genomic data, and large sensor networks.

Results

The project has fostered collaboration among researchers from GIPSA-lab, LJK, and TIMC-IMAG. Four doctoral dissertations co-supervised by researchers from two of the three different laboratories have been underway since October 2013. Persyvact has produced five articles and conference proceedings. The Persyvact project has organized numerous scientific events, including the Statlearn 2015 international workshop.

Seminars and meetings

• September 20, 2013 - Persyvact Kickoff Meeting
• October 17, 2013 - IXXI/Persyvact Seminar on Bayesian Approaches. More information
• November 19–20, 2013 – Persyvact Workshop: “Extremes and Copulas.” More information
• December 5, 2014 - First Persyvact interdisciplinary meeting between life scientists and applied mathematicians on human color perception. More information
• May 14, 2014 - SEMOVI/Persyvact Seminar on Machine Learning for Personalized Genomics.
• July 3, 2014 - One-day workshop on graphical models.
• November 13–14, 2014 – 2014 Astrostatical Days in Grenoble. More information
• March 31, 2014, and April 1, 2014 – Joint spring school (with the Khronos project team) on statistical inference.
• April 2–3, 2014 – Statlearn 2015 International Workshop on Statistical Learning.
• March 5, 2015 - ADM research initiative meeting. Slides on the Persyvact project. Slides by PhD student Alessandro Chiancone.

Publications

Mairal, J. (2015). Incremental majorization-minimization optimization with application to large-scale machine learning. SIAM Journal on Optimization, in press. HAL

Chiancone, A., Chanussot, J., & Girard, S. (2014). Collaborative Sliced Inverse Regression. Astrostatistics Conference. Grenoble, 2014 HAL

Clausel, M., Roueff, F., Taqqu, M., & Tudor, C.A. (2014). Asymptotic behavior of the quadratic variation of the sum of two Hermite processes of consecutive orders. Stochastic Processes and their Applications, 124: 2517–2541HAL

Duforet-Frebourg, N., & Blum, M. G. (2014). Non-stationary patterns of isolation-by-distance: inferring measures of local genetic differentiation using Bayesian kriging.Evolution,68(4), 1110-1123.HAL

Duforet-Frebourg, N., Bazin, E., & Blum, M. G. (2014). Genome scans for detecting footprints of local adaptation using a Bayesian factor model.Molecular Biology and Evolution, msu182.HAL

Frichot, E., Mathieu, F., Trouillon, T., Bouchard, G., & O. François (2014) Fast and efficient estimation of individual ancestry coefficients. Genetics 196: 973–983. HAL

He X., Condat L., Bioucas-Dias J., Chanussot J., & Xia J. (2014) A new pansharpening method based on spatial and spectral sparsity priors. IEEE Transactions on Image Processing, 23: 4160–4174. HAL

Mairal, J., Koniusz, P., Harchaoui, Z., & Schmid, C. (2014). Convolutional kernel networks. InAdvances in Neural Information Processing Systems(pp. 2627–2635). HAL

Prangle, D., Blum, M. G. B., Popovic, G., & Sisson, S. A. (2014). Diagnostic tools for approximate Bayesian computation using the coverage property.Australia and New Zealand Journal of Statistics, 56: 309–329. HAL

Clausel M, Roueff F, & Taqqu M (2013). The large-scale reduction principle and its application to hypothesis testing. Electronic Journal of Statistics, 9:153–203.HAL

Frichot, E., Schoville, S. D., Bouchard, G., & François, O. (2013). Testing for associations between loci and environmental gradients using latent factor mixed models.Molecular Biology and Evolution,30(7), 1687–1699.HAL

Members

• Sophie Achard (CNRS/Gipsa-lab)
• Hacheme Ayasso (UGA/Gipsa-lab)
• Michael Blum (CNRS/TIMC)
• Jean Marc Brossier (Grenoble INP/Gipsa-lab)
• Florent Chatelain (Grenoble INP/Gipsa-lab)
• Jocelyn Chanussot (Grenoble INP/Gipsa-lab)
• Marianne Clausel (UGA/LJK)
• Jean François Coeurjolly (UGA/LJK)
• Laurent Condat (CNRS/Gipsa-lab)
• Michel Desvignes (Grenoble INP/Gipsa-lab)
• Jean Baptiste Durand (Grenoble INP/Gipsa-lab)
• Florence Forbes (INRIA/LJK)
• Olivier François (Grenoble INP/TIMC)
• Stéphane Girard (INRIA/LJK)
• Radu Horaud (INRIA/LJK)
• Sophie Lambert (UGA/TIMC)
• Julien Mairal (INRIA/LJK)
• Marie-José Martinez (UGA/LJK)
• Olivier Michel, PR (Grenoble INP)
• Valérie Perrier (Grenoble INP/LJK)
• Nicolas Thierry-Mieg (CNRS/TIMC)

Coordinators

• Michael Blum (CNRS/TIMC)
• Marianne Clausel (UGA/LJK)
• Laurent Condat (CNRS/Gipsa-lab)