Cross-Layer Federated Learning for Lightweight IoT Intrusion Detection Systems.

With the proliferation of IoT devices, ensuring the security and privacy of these devices and their associated data has become a critical challenge. In this paper, we propose a federated sampling and lightweight intrusion-detection system for IoT networks that use K-meansfor sampling network traffic and identifying anomalies in a semi-supervised way. The system is designed to preserve data privacy by performing local clustering on each device and sharing only summary statistics with a central aggregator. The proposed system is particularly suitable for resource-constrained IoT devices such as sensors with limited computational and storage capabilities. We evaluate the system's performance using the publicly available NSL-KDD dataset. Our experiments and simulations demonstrate the effectiveness and efficiency of the proposed intrusion-detection system, highlighting the trade-offs between precision and recall when sharing statistics between workers and the coordinator. Notably, our experiments show that the proposed federated IDS can increase the true-positive rate up to 10% when the workers and the coordinator collaborate.

Deep learning methods for automatic evaluation of delayed enhancement-MRI. The results of the EMIDEC challenge.

A key factor for assessing the state of the heart after myocardial infarction (MI) is to measure whether the myocardium segment is viable after reperfusion or revascularization therapy. Delayed enhancement-MRI or DE-MRI, which is performed 10 min after injection of the contrast agent, provides high contrast between viable and nonviable myocardium and is therefore a method of choice to evaluate the extent of MI. To automatically assess myocardial status, the results of the EMIDEC challenge that focused on this task are presented in this paper. The challenge's main objectives were twofold. First, to evaluate if deep learning methods can distinguish between non-infarct and pathological exams, i.e. exams with or without hyperenhanced area. Second, to automatically calculate the extent of myocardial infarction. The publicly available database consists of 150 exams divided into 50 cases without any hyperenhanced area after injection of a contrast agent and 100 cases with myocardial infarction (and then with a hyperenhanced area on DE-MRI), whatever their inclusion in the cardiac emergency department. Along with MRI, clinical characteristics are also provided. The obtained results issued from several works show that the automatic classification of an exam is a reachable task (the best method providing an accuracy of 0.92), and the automatic segmentation of the myocardium is possible. However, the segmentation of the diseased area needs to be improved, mainly due to the small size of these areas and the lack of contrast with the surrounding structures.

Spatio-temporal dataset of building occupants.

The paper presents spatio-temporal dataset of building occupants captured using 200 Bluetooth Low Energy (BLE) beacons installed on different locations in two buildings. It contains 8426 data points of 11 building occupants collected with a sampling rate of 5 seconds during different times in a 12 days' interval. Each spatio-temporal data point comprises location and time components correspond to a building location which can be visualized using an OpenStreetMap (OSM) file of a building.

Embedded Implementation of VHR Satellite Image Segmentation.

Processing and analysis of Very High Resolution (VHR) satellite images provide a mass of crucial information, which can be used for urban planning, security issues or environmental monitoring. However, they are computationally expensive and, thus, time consuming, while some of the applications, such as natural disaster monitoring and prevention, require high efficiency performance. Fortunately, parallel computing techniques and embedded systems have made great progress in recent years, and a series of massively parallel image processing devices, such as digital signal processors or Field Programmable Gate Arrays (FPGAs), have been made available to engineers at a very convenient price and demonstrate significant advantages in terms of running-cost, embeddability, power consumption flexibility, etc. In this work, we designed a texture region segmentation method for very high resolution satellite images by using the level set algorithm and the multi-kernel theory in a high-abstraction C environment and realize its register-transfer level implementation with the help of a new proposed high-level synthesis-based design flow. The evaluation experiments demonstrate that the proposed design can produce high quality image segmentation with a significant running-cost advantage.

Architecture-Driven Level Set Optimization: From Clustering to Subpixel Image Segmentation.

Thanks to their effectiveness, active contour models (ACMs) are of great interest for computer vision scientists. The level set methods (LSMs) refer to the class of geometric active contours. Comparing with the other ACMs, in addition to subpixel accuracy, it has the intrinsic ability to automatically handle topological changes. Nevertheless, the LSMs are computationally expensive. A solution for their time consumption problem can be hardware acceleration using some massively parallel devices such as graphics processing units (GPUs). But the question is: which accuracy can we reach while still maintaining an adequate algorithm to massively parallel architecture? In this paper, we attempt to push back the compromise between, speed and accuracy, efficiency and effectiveness, to a higher level, comparing with state-of-the-art methods. To this end, we designed a novel architecture-aware hybrid central processing unit (CPU)-GPU LSM for image segmentation. The initialization step, using the well-known k -means algorithm, is fast although executed on a CPU, while the evolution equation of the active contour is inherently local and therefore suitable for GPU-based acceleration. The incorporation of local statistics in the level set evolution allowed our model to detect new boundaries which are not extracted by the used clustering algorithm. Comparing with some cutting-edge LSMs, the introduced model is faster, more accurate, less subject to giving local minima, and therefore suitable for automatic systems. Furthermore, it allows two-phase clustering algorithms to benefit from the numerous LSM advantages such as the ability to achieve robust and subpixel accurate segmentation results with smooth and closed contours. Intensive experiments demonstrate, objectively and subjectively, the good performance of the introduced framework both in terms of speed and accuracy.

Design, characterization and analysis of a 0.35 µm CMOS SPAD.

Most of the works about single-photon detectors rely on Single Photon Avalanche Diodes (SPADs) designed with dedicated technological processes in order to achieve single-photon sensitivity and excellent timing resolution. Instead, this paper focuses on the implementation of high-performance SPADs detectors manufactured in a standard 0.35-micron opto-CMOS technology provided by AMS. We propose a series of low-noise SPADs designed with a variable pitch from 20 µm down to 5 µm. This opens the further way to the integration of large arrays of optimized SPAD pixels with pitch of a few micrometers in order to provide high-resolution single-photon imagers. We experimentally demonstrate that a 20-micron SPAD appears as the most relevant detector in terms of Signal-to-Noise ratio, enabling emergence of large arrays of SPAD.

Multiple representations and mechanisms for visuomotor adaptation in young children.

In this study, we utilized transformed spatial mappings to perturb visuomotor integration in 5-yr-old children and adults. The participants were asked to perform pointing movements under five different conditions of visuomotor rotation (from 0° to 180°), which were designed to reveal explicit vs. implicit representations as well as the mechanisms underlying the visual-motor mapping. Several tests allowed us to separately evaluate sensorimotor (i.e., the dynamic dimension of movement) and cognitive (i.e., the explicit representations of target position and the strategies used by the participants) representations of visuo-proprioceptive distortion. Our results indicate that children do not establish representations in the same manner as adults and that children exhibit multiple visuomotor representations. Sensorimotor representations were relatively precise, presumably due to the recovery of proprioceptive information and efferent copy. Furthermore, a bidirectional mechanism was used to re-map visual and motor spaces. In contrast, cognitive representations were supplied with visual information and followed a unidirectional visual-motor mapping. Therefore, it appears that sensorimotor mechanisms develop before the use of explicit strategies during development, and young children showed impaired visuomotor adaptation when confronted with large distortions.

gSRT-Soft: a generic software application and some methodological guidelines to investigate implicit learning through visual-motor sequential tasks.

Serial reaction time tasks and, more generally, the visual-motor sequential paradigms are increasingly popular tools in a variety of research domains, from studies on implicit learning in laboratory contexts to the assessment of residual learning capabilities of patients in clinical settings. A consequence of this success, however, is the increased variability in paradigms and the difficulty inherent in respecting the methodological principles that two decades of experimental investigations have made more and more stringent. The purpose of the present article is to address those problems. We present a user-friendly application that simplifies running classical experiments, but is flexible enough to permit a broad range of nonstandard manipulations for more specific objectives. Basic methodological guidelines are also provided, as are suggestions for using the software to explore unconventional directions of research. The most recent version of gSRT-Soft may be obtained for free by contacting the authors.

Implicit learning of a repeated segment in continuous tracking: a reappraisal.

Several prior studies (e.g., Shea, Wulf, Whitacre, & Park, 2001; Wulf & Schmidt, 1997) have apparently demonstrated implicit learning of a repeated segment in continuous-tracking tasks. In two conceptual replications of these studies, we failed to reproduce the original findings. However, these findings were reproduced in a third experiment, in which we used the same repeated segment as that used in the Wulf et al. studies. Analyses of the velocity and the acceleration of the target suggests that this repeated segment could be easier to track than the random segments serving as control, accounting for the results of Wulf and collaborators. Overall these experiments suggest that learning a repeated segment in continuous-tracking tasks may be much more difficult than learning from a repeated sequence in conventional serial reaction time tasks. A possible explanation for this difference is outlined.