Hello !
This is my former academic website.
I am now co-founder at Owkin.

New email address: gilles.wainrib [at] owkin.com

Gilles Wainrib
Former assistant Professor at Département d'Informatique
team DATA, Ecole Normale Supérieure (Paris, France).
New email address: gilles.wainrib [at] owkin.com

Challenge Data

Challenge Data

  • > Machine learning challenges for teaching and research in data science.
  • > Promotes a free exchange of data and algorithmic knowledge, for education, science, industrial, social and medical applications.
  • > Challenges are supervised classification or regression problems, organized as competitions. Data provided by start-ups, innovative companies, medical centers and scientific experiments.

Recent preprints & publications

The asymptotic performance to linear echo-state networks
The Journal of Machine Learning Research 17 (2106) 1-35 (in press)
Romain Couillet, Gilles Wainrib, Hafiz Tiomoko Ali, Harry Sevi 

In this article, a study of the mean-square error (MSE) performance of linear echo-state neural networks is performed, both for training and testing tasks. Considering the realistic setting of noise present at the network nodes, we derive deterministic equivalents for the aforementioned MSE in the limit where the number of input data T and network size n both grow large. Specializing then the network connectivity matrix to specific random settings, we further obtain simple formulas that provide new insights on the performance of such networks.

A Random Matrix Approach to Echo-State Neural Networks (pdf)

Proceedings of the 33rd International Conference on Machine Learning, 2016.
Romain Couillet, Gilles Wainrib, Hafiz Tiomoko Ali, Harry Sevi

Recurrent neural networks, especially in their linear version, have provided many qualitative insights on their performance under different configurations. This article provides, through a novel random matrix framework, the quantitative counterpart of these performance results, specifically in the case of echo-state networks. Beyond mere insights, our approach conveys a deeper understanding on the core mechanism under play for both training and testing.

Colonic microRNA-based composite algorithm predicting drug responses in acute severe ulcerative colitis
Ian Morilla et. al, submitted

Acute severe ulcerative colitis (ASUC) is a severe condition that should be managed sequentially with intravenous steroids and, infliximab or cyclosporine in case of refractoriness. If medical treatment fails, salvage colectomy would be delayed and associated with excess mortality. Currently, there is no biomarker of drug response; the aim was to identify predictors of response to first and second-line treatments in ASUC. This study identified new biomarkers of response to treatment in patients with ASUC, enabling more accurate prediction of the benefit of therapy.

Context-dependent representation in recurrent neural networks
G. Wainrib, Preprint pdf

In order to assess the short-term memory performance of non-linear random neural networks, we introduce a measure to quantify the dependence of a neural representation upon the past context. We study this measure both numerically and theoretically using the mean-field theory for random neural networks, showing the existence of an optimal level of synaptic weights heterogeneity. We further investigate the influence of the network topology, in particular the symmetry of reciprocal synaptic connections, on this measure of context dependence, revealing the importance of considering the interplay between non-linearities and connectivity structure.

Ulcerative Colitis and smoking: an integrative network-based analysis of detoxification gene expression data
Y-P. Ding et al., to appear JCC

Inflammatory bowel diseases (IBD) including Crohn’s disease (CD) and ulcerative colitis (UC) are severe chronic intestinal disorders common in developed countries. Besides the well-characterized genetic contribution to IBD predisposition, environmental factors affect the incidence and medical history of IBD, among which active smoking has been shown as the most robust risk factor. The effect of smoking seems to be ambivalent since active smoking improves UC while it worsens CD. Although this clinical relationship between IBD and tobacco is well established, only a few experimental works have investigated the effect of smoking on the colonic barrier homeostasis focusing on xenobiotic detoxification genes. We performed a comprehensive and integrated comparative analysis of the global xenobiotic detoxification capacity of the normal colonic mucosa of healthy smokers and non-smokers versus the non-affected colonic mucosa of UC patients to improve our understanding of the colon susceptibility to environmental aggression. Among the 244 detoxification genes investigated, 65 were significantly dysregulated in UC patients, which corresponds to a specific disease signature. We then developed a network-based data analysis approach for differentially assessing changes in genetic interactions allowing identifying unexpected regulatory detoxification genes which could play a major role in the pathogenesis of UC or in the beneficial effect of smoking on the colonic mucosa of UC patients. These observations could help clinicians to better understand the protective effect of cigarette smoking in UC and will be useful to develop new therapeutic avenues and automated diagnostic strategies.

Branching random walks on binary strings and application to adaptive immunity
I. Balelli, V. Milisic and G. Wainrib, submitted [arxiv]

During the germinal center reaction, B lymphocytes proliferate, mutate and differentiate, while being submitted to a powerful selection, creating a micro-evolutionary mechanism at the heart of adaptive immunity. We introduce and analyze a simplified mathematical model of the division-mutation process, by considering random walks and branching random walks on graphs, whose structure reflects the associated mutation rules. In particular, we investigate how the combination of various division and mutation models influences the typical time-scales characterizing the efficiency of state space exploration for these processes, such as hitting times and cover times. Beyond the initial biological motivation, this framework is not limited to the modelling of B-cell learning process in germinal centers, as it may be relevant to model other evolutionary systems, but also information propagation in networks, gossip models or epidemic processes.

Network modeling of Crohn's disease
Jean-Marc Victor, Gaëlle Debret, Annick Lesne, Leigh Pascoe, Pascal Carrivain, Gilles Wainrib, Jean-Pierre Hugot, PLOS ONE 2016

Crohn's Disease (CD) is a complex genetic disorder related to genetic and environmental risk factors. We modelled the disease as a modular network of patho-physiological functions, each summarizing gene-environment interactions. The disease resulted from one or few specific combinations of the modules' functional states. Network aging dynamics was able to reproduce age-specific CD incidence curves and their variations over the past century in Western countries. The model allowed translating the Odds Ratios associated to at-risk alleles in terms of disease propensities of the functional modules. Finally, this modelling was successfully applied to other complex genetic disorders including ulcerative colitis, ankylosing spondylarthritis, multiple sclerosis and schizophrenia.

Research Interests

Probability theory / Dynamical systems

- limit theorems
- large deviations
- piecewise-deterministic Markov processes
- singular perturbations
- averaging principles
- stochastic bifurcations
- random matrix theory
- random walk on graphs
- partial differential equations
- random fields

Applications in theoretical biology / computer science

- Hodgkin Huxley models with stochastic ion channels
- Action potential generation and propagation
- Information transmission
- Noise-induced phenomena
- Synchronization
- Synaptic plasticity and learning
- Random neural networks
- Links with machine learning, reservoir computing
- Immune system modeling
- Gene regulatory networks

My collaborators

Vuk Milisic
Hatem Zaag

Contact information

Ecole Normale Supérieure
45 rue d'Ulm
75005 Paris

Email : gilles.wainrib [at] ens.fr

Recent Announcements

  • new affiliation -> new email !! my new e-mail adress is wainrib [at] math.univ-paris13.fr
    Posted Sep 20, 2011, 7:30 AM by Gilles Wainrib
  • New email adress My new email adress is:gwainrib [at] stanford.edu
    Posted Sep 8, 2010, 11:00 AM by Gilles Wainrib
Showing posts 1 - 2 of 2. View more »