André Smolarz

Multi-task learning with one-class SVM

Multi-task learning technologies have been developed to be an effective way to improve the generalization performance by training multiple related tasks simultaneously. The determination of the relatedness between tasks is usually the key to the formulation of a multi-task learning method. In this paper, we make the assumption that when tasks are related to each other, usually their models are close enough, that is, their models or their model parameters are close to a certain mean function. Following this task relatedness assumption, two multi-task learning formulations based on one-class support vector machines (one-class SVM) are presented. With the help of new kernel design, both multi-task learning methods can be solved by the optimization program of a single one-class SVM. Experiments conducted on both low-dimensional nonlinear toy dataset and high-dimensional textured images show that our approaches lead to very encouraging results.

Random feature subset selection in a nonstationary environment: Application to textured image segmentation

In this paper, we present a new feature subset selection method that intends to optimize or preserve the performances of a decisional system in case of nonstationary perturbations or loss of information. A two-step process is proposed. First, multiple classifiers are created based on random subspace method, and an initial decision is obtained by combining all the classifiers according to a weighted voting rule. Then, we classify anew all the observations with a subset of these classifiers, chosen in function of the quality of their related feature subspaces. To illustrate this approach, the two-class textured image segmentation problem is considered. Our attention is focused on trying to determine the optimum feature subsets in order to improve the classification accuracy at the borders between two textures. Experimental results demonstrate the effectiveness of the proposed approach.

Dynamic Feature Subspaces Selection for Decision in a Nonstationary Environment

The presence of noise, loss of information or feature nonstationarity in data is the limiting factor for many machine learning decision systems. Previous research has shown that relevant feature selection may be helpful to alleviate the impact of these possible perturbations. This paper presents a dynamical feature subspaces selection method based on ensembles of one-class Support Vector Machine (SVM), with the objective to optimize the performance of a decision system in such a nonstationary environment. Our method is predicated on the assumption that only the performance of the classifiers using perturbed features is degraded. We propose a mechanism for constructing an ensemble of classifiers based on a large number of feature subspaces generated from the initial full-dimensional space. In the phase of classification, the ensemble system is capable to select adaptively feature subspaces which are supposed to be immune to the nonstationary disturbance and to make the final decision by combining the individual decisions of classifiers built in these subspaces. One characteristic of this method is that we use the one-class SVM ensemble to accomplish simultaneously the tasks of feature subspace selection and classification. The effectiveness of the proposed method has been demonstrated through the experiments conducted in the context of textured image classification.

Feature subspaces selection via one-class SVM: Application to textured image segmentation

This paper presents a feature subspaces selection method which uses an ensemble of one-class SVMs. The objective is to improve or preserve the performance of a decision system in the presence of noise, loss of information or feature non-stationarity. The proposed method consists in first generating an ensemble of feature subspaces from the initial full-dimensional space, and then making the decision by using only the subspaces which are supposed to be immune to the non-stationary disturbance. One particularity of this method is that we use the one-class SVM ensemble to carry out the feature selection and the classification tasks at the same time. Textured image segmentation constitutes an appropriate application for the evaluation of the proposed approach. The experimental results demonstrate the effectiveness of the decision system that we have developed.

Dynamic Decision Method Based on Contextual Selection of Representation Subspaces

This paper presents a dynamical decision method derived from ensemble decision method. It is designed to be robust with respect to abrupt change of sensor response. Abrupt change may be caused by impulsive noise, sensor degradation or transmission fault in the case of an autonomous sensor network. It can also be caused by inconsistency of sensor responses due to local or sudden break of one monitored system property. The main idea is to divide the decision into several partial decisions and then to aggregate these to get the final one. The adaptation is the result of the aggregation process which aims at selecting and summarizing the partial decisions which are based on coherent information according to learnt models. The suggested method is presented. Experiments on a two-class image segmentation problem are performed and analyzed. The results assessed that the suggested method is more robust when an abrupt change occurs and is able to select efficiently the partial decision makers. This approach opens a wide field of applications and results are very encouraging.