Yann-Aël Le Borgne

Adaptive model selection for time series prediction in wireless sensor networks

In many practical applications of wireless sensor networks, the sensor nodes are required to report approximations of their readings at regular time intervals. For these applications, it has been shown that time series prediction techniques provide an effective way to reduce the communication effort while guaranteeing user-specified accuracy requirements on collected data. Achievable communication savings offered by time series prediction, however, strongly depend on the type of signal sensed, and in practice an inadequate a priori choice of a prediction model can lead to poor prediction performances. We propose in this paper the adaptive model selection algorithm, a lightweight, online algorithm that allows sensor nodes to autonomously determine a statistically good performing model among a set of candidate models. Experimental results obtained on the basis of 14 real-world sensor time series demonstrate the efficiency and versatility of the proposed framework in improving the communication savings.

New Routes from Minimal Approximation Error to Principal Components

We introduce two new methods of deriving the classical PCA in the framework of minimizing the mean square error upon performing a lower-dimensional approximation of the data. These methods are based on two forms of the mean square error function. One of the novelties of the presented methods is that the commonly employed process of subtraction of the mean of the data becomes part of the solution of the optimization problem and not a pre-analysis heuristic. We also derive the optimal basis and the minimum error of approximation in this framework and demonstrate the elegance of our solution in comparison with a recent solution in the framework.

Learned lessons in credit card fraud detection from a practitioner perspective

Abstract Billions of dollars of loss are caused every year due to fraudulent credit card transactions. The design of efficient fraud detection algorithms is key for reducing these losses, and more algorithms rely on advanced machine learning techniques to assist fraud investigators. The design of fraud detection algorithms is however particularly challenging due to non-stationary distribution of the data, highly imbalanced classes distributions and continuous streams of transactions. At the same time public data are scarcely available for confidentiality issues, leaving unanswered many questions about which is the best strategy to deal with them. In this paper we provide some answers from the practitioner’s perspective by focusing on three crucial issues: unbalancedness, non-stationarity and assessment. The analysis is made possible by a real credit card dataset provided by our industrial partner.

Distributed principal component analysis for wireless sensor networks

The Principal Component Analysis (PCA) is a data dimensionality reduction tech-nique well-suited for processing data from sensor networks. It can be applied to tasks like compression, event detection, and event recognition. This technique is based on a linear trans-form where the sensor measurements are projected on a set of principal components. When sensor measurements are correlated, a small set of principal components can explain most of the measurements variability. This allows to significantly decrease the amount of radio communication and of energy consumption. In this paper, we show that the power iteration method can be distributed in a sensor network in order to compute an approximation of the principal components. The proposed implementation relies on an aggregation service, which has recently been shown to provide a suitable framework for distributing the computation of a linear transform within a sensor network. We also extend this previous work by providing a detailed analysis of the computational, memory, and communication costs involved. A com-pression experiment involving real data validates the algorithm and illustrates the tradeoffs between accuracy and communication costs.

Machine Learning Strategies for Time Series Forecasting

The increasing availability of large amounts of historical data and the need of performing accurate forecasting of future behavior in several scientific and applied domains demands the definition of robust and efficient techniques able to infer from observations the stochastic dependency between past and future. The forecasting domain has been influenced, from the 1960s on, by linear statistical methods such as ARIMA models. More recently, machine learning models have drawn attention and have established themselves as serious contenders to classical statistical models in the forecasting community. This chapter presents an overview of machine learning techniques in time series forecasting by focusing on three aspects: the formalization of one-step forecasting problems as supervised learning tasks, the discussion of local learning techniques as an effective tool for dealing with temporal data and the role of the forecasting strategy when we move from one-step to multiple-step forecasting.

Decentralised reinforcement learning for energy-efficient scheduling in wireless sensor networks

We present a self-organising reinforcement learning (RL) approach for scheduling the wake-up cycles of nodes in a wireless sensor network. The approach is fully decentralised, and allows sensor nodes to schedule their active periods based only on their interactions with neighbouring nodes. Compared to standard scheduling mechanisms such as SMAC, the benefits of the proposed approach are twofold. First, the nodes do not need to synchronise explicitly, since synchronisation is achieved by the successful exchange of data messages in the data collection process. Second, the learning process allows nodes competing for the radio channel to desynchronise in such a way that radio interferences and therefore packet collisions are significantly reduced. This results in shorter communication schedules, allowing to not only reduce energy consumption by reducing the wake-up cycles of sensor nodes, but also to decrease the data retrieval latency. We implement this RL approach in the OMNET++ sensor network simulator, and illustrate how sensor nodes arranged in line, mesh and grid topologies autonomously uncover schedules that favour the successful delivery of messages along a routing tree while avoiding interferences.

Une plate-forme technologique liée à la paralysie cérébrale - Le projet ICT4Rehab

The musculoskeletal system (MSS) is essential to allow us performing every-day tasks, being able to have a professional life or developing social interactions with our entourage. MSS pathologies have a significant impact on our daily life. It is therefore not surprising to find MSS-related health problems at the top of global statistics on professional absenteeism or societal health costs. The MSS is also involved in central nervous conditions, such as cerebral palsy (CP). Such conditions show complex etiology that complicates the interpretation of the observable clinical signs and the establishment of a wide consensus on the best practices to adopt for clinical monitoring and patient follow-up. These elements justify the organization of fundamental and applied research projects aiming to develop new methods to help clinicians to cope with the complexity of some MSS disorders. The ICT4Rehab project (www.ict4rehab.org) developed an integrated platform providing tools that enable easier management and visualization of clinical information related to the MSS of CP patients. This platform is opened to every interested clinical centre.

Simulation architecture for data processing algorithms in wireless sensor networks

Wireless sensor networks, by providing an unprecedented way of interacting with the physical environment, have become a hot topic for research over the last few years. As with any new technology, results from real experimentations using these networks are still scarce, as real deployments are either costly, or still unfeasible in the current state of technology. There is therefore an increasing need for simulation tools allowing the testing of different architectures, communication protocols or information processing algorithms in sensor networks. In this paper, we investigate a simulation framework for the testing of data processing in wireless sensor network applications. In a first stage, data is generated using partial differential equations, allowing the modeling of a large panel of physical phenomena. In a second stage, sensing unit operating system and network constraints are simulated using an instance of a versatile simulator to account for the platform characteristics. Insights provided by the proposed simulation frame are illustrated by a set of experiments on a heat source detection task.

Reinforcement Learning for Self-organizing Wake-Up Scheduling in Wireless Sensor Networks

Wake-up scheduling is a challenging problem in wireless sensor networks. It was recently shown that a promising approach for solving this problem is to rely on reinforcement learning (RL). The RL approach is particularly attractive since it allows the sensor nodes to coordinate through local interactions alone, without the need of central mediator or any form of explicit coordination. This article extends previous work by experimentally studying the behavior of RL wake-up scheduling on a set of three different network topologies, namely line, mesh and grid topologies. The experiments are run using OMNET++, a the state-of-the-art network simulator. The obtained results show how simple and computationally bounded sensor nodes are able to coordinate their wake-up cycles in a distributed way in order to improve the global system performance. The main insight of these experiments is to show that sensor nodes learn to synchronize if they have to cooperate for forwarding data, and learn to desynchronize in order to avoid interferences. This synchronization/desynchronization behavior, referred to for short as (de)synchronicity, allows to improve the message throughput even for very low duty cycles.

SCARFF: a Scalable Framework for Streaming Credit Card Fraud Detection with Spark

The expansion of the electronic commerce, together with an increasing confidence of customers in electronic payments, makes of fraud detection a critical factor. Detecting frauds in (nearly) real time setting demands the design and the implementation of scalable learning techniques able to ingest and analyse massive amounts of streaming data. Recent advances in analytics and the availability of open source solutions for Big Data storage and processing open new perspectives to the fraud detection field. In this paper we present a SCAlable Real-time Fraud Finder (SCARFF) which integrates Big Data tools (Kafka, Spark and Cassandra) with a machine learning approach which deals with imbalance, nonstationarity and feedback latency. Experimental results on a massive dataset of real credit card transactions show that this framework is scalable, efficient and accurate over a big stream of transactions.