Machine learning techniques, such as deep learning and ensemble methods, are widely used in various domains due to their ability to handle complex real-world tasks. However, their black-box nature has raised multiple concerns about the fairness, trustworthiness, and transparency of computer-assisted decision-making. This has led to the emergence of local post-hoc explainability methods, which offer explanations for individual decisions made by black-box algorithms. Among these methods, Kernel SHAP is widely used due to its model-agnostic nature and its well-founded theoretical framework. Despite these strengths, Kernel SHAP suffers from high instability: different executions of the method with the same inputs can lead to significantly different explanations, which diminishes the relevance of the explanations. The contribution of this paper is two-fold. On the one hand, we show that Kernel SHAP's instability is caused by its stochastic neighbor selection procedure, which we adapt to achieve full stability without compromising explanation fidelity. On the other hand, we show that by restricting the neighbors generation to perturbations of size 1 -- which we call the coalitions of Layer 1 -- we obtain a novel feature-attribution method that is fully stable, computationally efficient, and still meaningful.

Real-time monitoring of hematophagous diptera (such as mosquitoes) populations in the field is a crucial challenge to foresee vaccination campaigns and to restrain potential diseases spreading. However, current methods heavily rely on costly DNA extraction which is destructive, costly, time consuming and requires experts. The contributions of this work are: 1) the usage of a new type of imaging, named Wing Interference Patterns (WIPs), which is non-destructive and easier to produce during in the field experiments; 2) a deep learning architecture which is optimized for very low computation cost, memory usage and a short inference time; 3) the use of a dataset of more than 50 medically important species of hematophagous diptera with more than 3000 images of WIPs. With these contributions, we demonstrate that WIPs are an excellent medium to automatically recognize a large amount of hematophagous diptera species with very high accuracy and low computational cost convolutional neural network.

International audience

Most of Wireless Sensor Networks are deployed to monitor a set of targets over a specified area. The lifetime of such a network is defined as the time duration from the network deployment till the time when one target is no longer covered. Thus, this lifetime is limited by the energy resource of sensor nodes. In order to maximize the lifetime of the network, only a subset of nodes capable of covering all targets are activated at a time while the others are put in sleep mode to save their energy. When an active sensor fails, a recovery procedure should be executed to keep all targets covered. In this paper we propose a new self-healing method for network reconfiguration in case of failure of an active node. Simulation results show that this method increases the network dependability by reducing the network unavailability time up to 90 % % compared to the dynamic maintenance for networks with more than 200 sensor nodes of sensing range equals to 10, uniformly deployed over a 50 × 50 square to cover 50 targets also uniformly deployed over the same area.

Prediction of user behavior in Social Networks is important for a lot of applications, ranging from marketing to social community management. In this paper, we develop and test a model to estimate the propensity of a user to stop using the social platform in a near future. This problem is called churn prediction and has been extensively studied in telecommunication networks. We focus here on building a statistical model estimating the probability that a user will leave the social network in the near future. The model is based on graph attributes extracted in the user's vicinity. We present a novel algorithm to accurately detect overlapping local communities in social graphs. Our algorithm outperforms the state of the art methods and is able to deal with pathological cases which can occur in real networks. We show that using attributes computed from the local community around the user allows to build a robust statistical model to predict churn. Our ideas are tested on one of the largest French social blog platform, Sky rock, where millions of teenagers interact daily.

Nowadays, enterprises can provide computing services through their intranet networks by letting their available resources be used as virtual clusters for scientific computation during idle periods such as nights, weekends, and holidays. Generally, these idle periods do not permit to carry out the computations completely. It is therefore necessary to save the context of uncompleted applications for possible restart. This checkpointing mechanism is subject to resource constraints: the network bandwidth, the disk bandwidth, and the delay T imposed for releasing the workstations. We first introduce a function bw that gives the bandwidth bw(m,V) of a system during the checkpointing of m applications with aggregated memory requirement V. Assuming that this bandwidth is shared equitably among the applications, the scheduling problem becomes a sequence of knapsack problems with nonlinear constraints for which we propose approximate solutions. Experiments carried out on Grid5000 show that the running time of this algorithm is negligible compared to the delay T which is of the order of few minutes. This means that the proposed scheduling algorithm does not induce a significant overhead on the checkpointing process. As a consequence, our mechanism can be incorporated in a batch scheduler.

We consider a context where the available resources of the Intranet of a company are used as a virtual cluster for scientific computation, during the idle periods (nights, weekends, holidays, ). Generally, these idle periods do not permit to carry out completely the computations. For instance, a workstation mobilized during the night must be released in the morning to make it available for the employee, even if the application running on it is not completed. It is therefore necessary to save the context of uncompleted applications for possible restart. Hereafter, we assume that the computations running on the workstations are independent from each other. The checkpointing mechanism which ensures the continuity of applications is subject to resource constraints : the network bandwidth, the disk bandwidth and the delay T imposed for releasing the workstations. We first show that the designing of a scheduling strategy which optimizes resource consumption while taking into account the above constraints, can be formalized as a variant of the classical 0/1 knapsack problem. We then propose an algorithm whose implementation does not have a significant overhead on checkpointing mechanisms. Experiments carried out on a real cluster show that this algorithm performs better than the naive scheduling algorithm which selects the applications one after the other in order of decreasing amount of resource consumption.

In this paper, we study the QR factorization of a square matrix of order n, through Givens rotations. The task allocation method developed among others by QUINTON, SAKHO and TCHUENTE, is used to parallelize the QR factorization scheme introduced by SAMEH and KUCK. This yields a solution which runs in optimal time T* = 3n-1, on an hexagonally connected array of size 5n2/18 + 0(n), composed of programmable cells. This improves the best previously known solution introduced by COSNARD and ROBERT, and which runs in optimal time on an array of size 3n2/8 + 0(n).

This correspondence studies the time complexity of the parallel computation of the product C = A.B of two dense square matrices A, B of order n, on a class of rectangular orthogonally connected systolic arrays, which are the two-dimensional extensions of the classical pipeline scheme. Such arrays are composed of multiply-add cells without local memory, and, as C is computed, the coefficients cij move vertically, whereas aik and bkj move horizontally in opposite directions. We first introduce a combinatorial formulation of the problem. Then we show that, if the cycle-time of a multiply-add cell is taken as time unit, and if T(p, m) denotes the running time of an optimal algorithm associated with an array of size p x m, then Minpm T(p,m) = 3n -2, and the minimum value of p.m for which this bound is tight is n.n [resp. n(n + 1)] if n is odd (resp. even). When compared to the algorithms previously proposed for the class of arrays based on cells without local memory, the solutions exhibited here appear to be the best, because they are the only ones which run in time T < = 3n -2 on a network of size S < = n(n + 1).