Rosa Altilio

Multimedia and Gaming Technologies for Telerehabilitation of Motor Disabilities [Leading Edge]

Rehabilitation for chronic conditions resulting from acute or progressive disease might be delivered in an outpatient facility as in the case of telerehabilitation, self rehabilitation and, more generally, in the context of home-based rehabilitation to improve the patients’ quality of life. Here we present the emerging field of home-based applications for continuous digital health, focusing in particular on low-cost rehabilitation systems for motor disabilities based on multimedia and gaming technologies. Innovative technologies for telerehabilitation are illustrated. We also present recent advances in telerehabilitation, considering the most relevant projects that best represent new trends for research and development of new technologies and applications in this context.

Embedding of time series for the prediction in photovoltaic power plants

The ability to forecast the power produced by renewable energy plants in short and middle terms is a key issue to allow an high-level penetration of the distributed generation into the grid infrastructure. Forecasting energy production is mandatory for dispatching and distribution issues, at the Trasmission System Operator level, as well as electrical distributors and power system operators level. In this paper we present three techniques based on neural and fuzzy neural networks, which are well suited to predict data sequences stemming from real-world applications. The preliminary results concerning the prediction of the power generated by a large-scale photovoltaic plant in Italy confirm the reliability and accuracy of the proposed approaches.

Recent Advances on Distributed Unsupervised Learning

Distributed machine learning is a problem of inferring a desired relation when the training data is distributed throughout a network of agents (e.g. sensor networks, robot swarms, etc.). A typical problem of unsupervised learning is clustering, that is grouping patterns based on some similarity/dissimilarity measures. Provided they are highly scalable, fault-tolerant and energy efficient, clustering algorithms can be adopted in large-scale distributed systems. This work surveys the state-of-the-art in this field, presenting algorithms that solve the distributed clustering problem efficiently, with particular attention to the computation and clustering criteria.

Privacy-Preserving Data Mining for Distributed Medical Scenarios

In this paper, we consider the application of data mining methods in medical contexts, wherein the data to be analysed (e.g. records from different patients) is distributed among multiple clinical parties. Although inference procedures could provide meaningful medical information (such as optimal clustering of the subjects), each party is forbidden to disclose its local dataset to a centralized location, due to privacy concerns over sensible portions of the dataset. To this end, we propose a general framework enabling the parties involved to perform (in a decentralized fashion) any data mining procedure relying solely on the Euclidean distance among patterns, including kernel methods, spectral clustering, and so on. Specifically, the problem is recast as a decentralized matrix completion problem, whose proposed solution does not require the presence of a centralized coordinator, and full privacy of the original data can be ensured by the use of different strategies, including random multiplicative updates for secure computation of distances. Experimental results support our proposal as an efficient tool for performing clustering and classification in distributed medical contexts. As an example, on the known Pima Indians Diabetes dataset, we obtain a Rand-Index for clustering of 0.52 against 0.54 of the (unfeasible) centralized solution, while on the Parkinson speech database we increase from 0.45 to 0.50.

Finite precision implementation of random vector functional-link networks

The increasing amount of data to be processed coming from multiple sources, as in the case of sensor networks, and the need to cope with constraints of security and privacy, make necessary the use of computationally efficient techniques on simple and cheap hardware architectures often distributed in pervasive scenarios. Random Vector Functional-Link is a neural network model usually adopted for processing distributed big data, but no constraints have been considered so far to deal with limited hardware resources. This paper is focused on implementing a modified version of the Random Vector Functional-Link network with finite precision arithmetic, in order to make it suited to hardware architectures even based on a simple microcontroller. A genetic optimization is also proposed to ensure that the overall performance is comparable with standard software implementations. The numerical results prove the efficacy of the proposed approach.

Distributed on-line learning for random-weight fuzzy neural networks

The Random-Weight Fuzzy Neural Network is an inference system where the fuzzy rule parameters of antecedents (i.e., membership functions) are randomly generated and the ones of consequents are estimated using a Regularized Least Squares algorithm. In this regard, we propose an on-line learning algorithm under the hypothesis of training data distributed across a network of interconnected agents. In particular, we assume that each agent in the network receives a stream of data as a sequence of mini-batches. When receiving a new chunk of data, each agent updates its estimate of the consequent parameters and, periodically, all agents agree on a common model through the Distributed Average Consensus protocol. The learning algorithm is faster than a solution based on a centralized training set and it does not rely on any coordination authority. The experimental results on well-known datasets validate our proposal.

A nonuniform quantizer for hardware implementation of neural networks

New trends in neural computation, now dealing with distributed learning on pervasive sensor networks and multiple sources of big data, make necessary the use of computationally efficient techniques to be implemented on simple and cheap hardware architectures. In this paper, a nonuniform quantization at the input layer of neural networks is introduced, in order to optimize their implementation on hardware architectures based on a finite precision arithmetic. Namely, we propose a nonlinear A/D conversion of input signals by considering the actual structure of data to be processed. Random Vector Functional-Link is considered as the reference model for neural networks and a genetic optimization is adopted for determining the quantization levels to be found. The proposed approach is assessed by several experimental results obtained on well-known benchmarks for the general problem of data regression.

A genetic algorithm for feature selection in gait analysis

This paper deals with the opportunity of extracting useful information from medical data retrieved directly from a stereophotogrammetric system applied to gait analysis, which aims at controlling movements of patients affected by neurological diseases. The proposed approach is intended to a feature selection procedure as an optimization strategy based on genetic algorithms, where the misclassification error of healthy/diseased patients is adopted as the fitness function. This procedure will be used for estimating the performance of widely used classification algorithms, whose performance has been ascertained in many real-world problems with respect to well-known classification benchmarks, both in terms of number of selected features and classification accuracy. Moreover, the technique herein described will provide a useful tool in the context of medical diagnosis. In fact, we will prove that for the classification problem at hand the whole set of features is redundant and it can be significantly pruned. The obtained results on a real dataset acquired in our biomechanics laboratory show a very interesting classification accuracy using six features only among the sixteen acquired by the stereophotogrammetric system.

A Classification Approach to Modeling Financial Time Series

In this paper, several classification methods are applied for modeling financial time series with the aim to predict the trend of successive prices. By using a suitable embedding technique, a pattern of past prices is assigned a class if the variation of the next price is over, under or stable with respect to a given threshold. Furthermore, a sensitivity analysis is performed in order to verify if the value of such a threshold influences the prediction accuracy. The experimental results on the case study of WTI crude oil commodity show a good classification accuracy of the next (predicted) trend, and the best performance is achieved by the K-Nearest Neighbors classification strategy.

Takagi-sugeno fuzzy systems applied to voltage prediction of photovoltaic plants

High penetration level of intermittent and variable renewable electricity generation introduces signicant challenges to energy management of modern smart grids. Solar photo-voltaics and wind energy have uncertain and non-dispatchable output which leads to concerns regarding the technical and economic feasibility of a reliable integration of large amounts of variable generation into electric grids. In this scenario, accurate forecasting of renewable generation outputs is of paramount importance to secure operation of smart grids. In this paper, we present a study on the use of fuzzy neural networks and their application to the prediction of solar photovoltaic outputs. The new learning strategy is suited to any fuzzy inference model. The comparison with respect to well-known neural and fuzzy neural models will prove that our approach is able to follow the behavior of the underlying unknown process with a good prediction of the observed time series.