-
Statistical learning with Lipschitz and convex loss functions
Authors:
Geoffrey Chinot,
Lecué Guillaume,
Lerasle Matthieu
Abstract:
We obtain risk bounds for Empirical Risk Minimizers (ERM) and minmax Median-Of-Means (MOM) estimators based on loss functions that are both Lipschitz and convex. Results for the ERM are derived without assumptions on the outputs and under subgaussian assumptions on the design and a new "local Bernstein assumption" on the class of predictors. Similar results are shown for minmax MOM estimators in a…
▽ More
We obtain risk bounds for Empirical Risk Minimizers (ERM) and minmax Median-Of-Means (MOM) estimators based on loss functions that are both Lipschitz and convex. Results for the ERM are derived without assumptions on the outputs and under subgaussian assumptions on the design and a new "local Bernstein assumption" on the class of predictors. Similar results are shown for minmax MOM estimators in a close setting where the design is only supposed to satisfy moment assumptions, relaxing the Subgaussian hypothesis necessary for ERM. The analysis of minmax MOM estimators is not based on the small ball assumption (SBA) as it was the case in the first analysis of minmax MOM estimators. In particular, the basic example of non parametric statistics where the learning class is the linear span of localized bases, that does not satisfy SBA can now be handled. Finally, minmax MOM estimators are analysed in a setting where the local Bernstein condition is also dropped out. It is shown to achieve an oracle inequality with exponentially large probability under minimal assumptions insuring the existence of all objects.
△ Less
Submitted 28 June, 2019; v1 submitted 2 October, 2018;
originally announced October 2018.
-
Tracking bitcoin users activity using community detection on a network of weak signals
Authors:
Remy Cazabet,
Baccour Rym,
Latapy Matthieu,
Cazabet Remy
Abstract:
Bitcoin is a cryptocurrency attracting a lot of interest both from the general public and researchers. There is an ongoing debate on the question of users' anonymity: while the Bitcoin protocol has been designed to ensure that the activity of individual users could not be tracked, some methods have been proposed to partially bypass this limitation. In this article, we show how the Bitcoin transact…
▽ More
Bitcoin is a cryptocurrency attracting a lot of interest both from the general public and researchers. There is an ongoing debate on the question of users' anonymity: while the Bitcoin protocol has been designed to ensure that the activity of individual users could not be tracked, some methods have been proposed to partially bypass this limitation. In this article, we show how the Bitcoin transaction network can be studied using complex networks analysis techniques, and in particular how community detection can be efficiently used to re-identify multiple addresses belonging to a same user.
△ Less
Submitted 23 October, 2017;
originally announced October 2017.
-
Learning from MOM's principles: Le Cam's approach
Authors:
Lecué Guillaume,
Lerasle Matthieu
Abstract:
We obtain estimation error rates for estimators obtained by aggregation of regularized median-of-means tests, following a construction of Le Cam. The results hold with exponentially large probability -- as in the gaussian framework with independent noise- under only weak moments assumptions on data and without assuming independence between noise and design. Any norm may be used for regularization.…
▽ More
We obtain estimation error rates for estimators obtained by aggregation of regularized median-of-means tests, following a construction of Le Cam. The results hold with exponentially large probability -- as in the gaussian framework with independent noise- under only weak moments assumptions on data and without assuming independence between noise and design. Any norm may be used for regularization. When it has some sparsity inducing power we recover sparse rates of convergence.
The procedure is robust since a large part of data may be corrupted, these outliers have nothing to do with the oracle we want to reconstruct. Our general risk bound is of order \begin{equation*} \max\left(\mbox{minimax rate in the i.i.d. setup}, \frac{\text{number of outliers}}{\text{number of observations}}\right) \enspace. \end{equation*}In particular, the number of outliers may be as large as (number of data) $\times$(minimax rate) without affecting this rate. The other data do not have to be identically distributed but should only have equivalent $L^1$ and $L^2$ moments.
For example, the minimax rate $s \log(ed/s)/N$ of recovery of a $s$-sparse vector in $\mathbb{R}^d$ is achieved with exponentially large probability by a median-of-means version of the LASSO when the noise has $q_0$ moments for some $q_0>2$, the entries of the design matrix should have $C_0\log(ed)$ moments and the dataset can be corrupted up to $C_1 s \log(ed/s)$ outliers.
△ Less
Submitted 18 July, 2017; v1 submitted 8 January, 2017;
originally announced January 2017.
-
Rapid onset of collectivity in the vicinity of 78Ni
Authors:
Lebois Matthieu,
David Verney,
Fadi Ibrahim,
Said Essabaa,
Faiçal Azaiez,
Maher Cheikh Mhamed,
Evelyne Cottereau,
Cuong Phan Viet,
Mathieu Ferraton,
Kieran Flanagan,
Serge Franchoo,
Dominique Guillemaud Mueller,
Fairouz Hammache,
Christophe Lau,
François Le Blanc,
Jean François Le Du,
Baptiste Mouginot,
Costel Petrache,
Brigitte Roussière,
Lionel Sagui,
Nicolas De Sereville,
Iulian Stefan,
Benoit Tastet
Abstract:
gamma-rays following the B and B-n decay of the very neutron rich 84Ga produced by photo-fission of 238U have been studied at the newly built ISOL facility of IPN Orsay: ALTO. Two activities were observed and assigned to two B-decaying states: 84gGa, I = (0\^-) and 84mGa, I = (3\^-, 4\^-). Excitation energies of the 2+1 and 4+1 excited states of 84Ge were measured at E(2+1) = 624.3 keV and E(4+1…
▽ More
gamma-rays following the B and B-n decay of the very neutron rich 84Ga produced by photo-fission of 238U have been studied at the newly built ISOL facility of IPN Orsay: ALTO. Two activities were observed and assigned to two B-decaying states: 84gGa, I = (0\^-) and 84mGa, I = (3\^-, 4\^-). Excitation energies of the 2+1 and 4+1 excited states of 84Ge were measured at E(2+1) = 624.3 keV and E(4+1) = 1670.1 keV. Comparison with HFB+GCM calculations allows to establish the collective character of this nucleus indicating a substantial N=50 core polarization. The excitation energy of the 1/2+1 state in 83Ga known to carry a large part of the neutron 3s1/2 strength was measured at 247.8keV. Altogether these data allow to confirm the new single particle state ordering which appears immediately after the double Z=28 and N=50 shell closure and to designate 78Ni as a fragile and easily polarized doubly-magic core.
△ Less
Submitted 21 October, 2008;
originally announced October 2008.
