Francisco Javier González-Serrano

Median equivariant adaptive separation via independence: application to communications

Abstract The authors propose a family of algorithms to solve the blind source separation (BSS) problem, focusing on digital communications. We first include a revision of the equivariant adaptive separation via independence (EASI). These results provide a better characterization of the stability condition. Starting from this analysis we propose a new method, the Median EASI, whose stability condition presents useful properties. Next, we prove this method to be more noise-robust for some sources such as digitally modulated signals. Some experiments are included to show how the Median EASI allows phase recovering, an important issue in communications. Finally, we propose the median EASI as a blind asynchronous CDMA multiuser detector.

Comparison of optimization algorithms in the sensor selection for predictive target tracking

Abstract This paper addresses the selection of sensors for target localization and tracking under nonlinear and nonGaussian dynamic conditions. We have used the Posterior Cramer-Rao lower Bound (PCRB) as the performance-based optimization criteria because of its built-in capability to produce online estimation performance predictions, a “must” for high maneuverable targets or when slow-response sensors are used. In this paper, we analyze, and compare, three optimization algorithms: genetic algorithm (GA), particle swarm optimization (PSO), and a new discrete-variant of the cuckoo search algorithm (CS). Finally, we propose local-search versions of the previous optimization algorithms that provide a significant reduction of the computation time.

Adaptive blind joint source-phase separation in digital communications

The authors present a set of adaptive algorithms for the blind separation of independent sources (BSS). Source separation consists of recovering a set of independent signals from some linear instant mixtures of them, the coefficients of the mixing matrix being unknown. The relative (or natural) gradient has been widely used in these problems. We propose to replace it by the median gradient. We extend the concept to the complex case to cope with digitally modulated signals. It results in a new family of methods with better performance and phase recovering properties. Examples include separation of instant mixtures of different communication signals to illustrate the solutions proposed. As a new result we apply the proposed algorithms to asynchronous CDMA.

Training Support Vector Machines with privacy-protected data

Abstract In this paper, we address a machine learning task using encrypted training data. Our basic scenario has three parties: Data Owners, who own private data; an Application, which wants to train and use an arbitrary machine learning model on the Users’ data; and an Authorization Server, which provides Data Owners with public and secret keys of a partial homomorphic cryptosystem (that protects the privacy of their data), authorizes the Application to get access to the encrypted data, and assists it in those computations not supported by the partial homomorphism. As machine learning model, we have selected the Support Vector Machine (SVM) due to its excellent performance in supervised classification tasks. We evaluate two well known SVM algorithms, and we also propose a new semiparametric SVM scheme better suited for the privacy-protected scenario. At the end of the paper, a performance analysis regarding the accuracy and the complexity of the developed algorithms and protocols is presented.

Radial Basis Function Interpolation for Signal-Model-Independent Localization

In this paper, we propose a novel localization algorithm to be used in applications where the measurement model is neither accurate nor complete. In our algorithm, we apply radial basis function (RBF) interpolation to evaluate the measurement function on the entire surveillance area and, then, estimate the target position. Since the signal function is sparse in the spatial domain, we also propose to use sparse optimization techniques (LASSO) both to efficiently compute the weights for the RBF and to improve the interpolated function quality. Simulation results show good performance in the localization of single and multiple targets.

State estimation using an extended Kalman filter with privacy-protected observed inputs

In this paper, we focus on the parameter estimation of dynamic state-space models using privacy-protected data. We consider an scenario with two parties: on one side, the data owner, which provides privacy-protected observations to, on the other side, an algorithm owner, that processes them to learn the systems state vector. We combine additive homomorphic encryption and Secure Multiparty Computation protocols to develop secure functions (multiplication, division, matrix inversion) that keep all the intermediate values encrypted in order to effectively preserve the data privacy. As an application, we consider a tracking problem, in which a Extended Kalman Filter estimates the position, velocity and acceleration of a moving target in a collaborative environment where encrypted distance measurements are used.

Adaptive blind multiuser detection in asynchronous CDMA

This paper applies blind source separation (BSS) to the design of an adaptive blind multiuser detector for an asynchronous code division multiple access (CDMA) system. Our starting hypotheses are the spreading factor and the number of users transmitting in the system. We assume too that we do not have knowledge of the spreading sequences or the delays affecting each of the users in the system. The proposed blind receptor has a similar performance compared to the synchronized matched filter.

Supervised machine learning using encrypted training data

Preservation of privacy in data mining and machine learning has emerged as an absolute prerequisite in many practical scenarios, especially when the processing of sensitive data is outsourced to an external third party. Currently, privacy preservation methods are mainly based on randomization and/or perturbation, secure multiparty computations and cryptographic methods. In this paper, we take advantage of the partial homomorphic property of some cryptosystems to train simple machine learning models with encrypted data. Our basic scenario has three parties: multiple Data Owners, which provide encrypted training examples; the Algorithm Owner (or Application), which processes them to adjust the parameters of its models; and a semi-trusted third party, which provides privacy and secure computation services to the Application in some operations not supported by the homomorphic cryptosystem. In particular, we focus on two issues: the use of multiple-key cryptosystems, and the impact of the quantization of real-valued input data required before encryption. In addition, we develop primitives based on the outsourcing of a reduced set of operations that allows to implement general machine learning algorithms using efficient dedicated hardware. As applications, we consider the training of classifiers using privacy-protected data and the tracking of a moving target using encrypted distance measurements.

Generalized CMAC adaptive ensembles for concept-drifting data streams

In this paper we propose to use an adaptive ensemble learning framework with different levels of diversity to handle streams of data in non-stationary scenarios in which concept drifts are present. Our adaptive system consists of two ensembles, each one with a different level of diversity (from high to low), and, therefore, with different and complementary capabilities, that are adaptively combined to obtain an overall system of improved performance. In our approach, the ensemble members are generalized CMACs, a linear-in-the-parameters network. The ensemble of CMACs provides a reasonable trade-off between expressive power, simplicity, and fast learning speed. At the end of the paper, we provide a performance analysis of the proposed learning framework on benchmark datasets with concept drifts of different levels of severity and speed.