Publications

2020

• Efficient Scheduling of FPGAs for Cloud Data Center Infrastructures
  
  • Bertolino Matteo
  • Enrici Andrea
  • Pacalet Renaud
  • Apvrille Ludovic
  , 2020. In modern cloud data centers, reconfigurable devices can be directly connected to a data center's network. This configuration enables FPGAs to be rented for acceleration of data-intensive workloads. In this context, novel scheduling solutions are needed to maximize the utilization (profitability) of FPGAs, e.g., reduce latency and resource fragmentation. Algorithms that schedule groups of tasks (clusters, packs), rather than individual tasks (list scheduling), well match the functioning of FPGAs. Here, groups of tasks that execute together are interposed by hardware reconfigurations. In this paper, we propose a heuristic based on a novel method for grouping tasks. These are gathered around a high-latency task that hides the latency of remaining tasks within the same group. We evaluated our solution on a benchmark of almost 30000 random workloads, synthesized from realistic designs (i.e., topology, resource occupancy). For this testbench, on average, our heuristic produces optimum makespan solutions in 71.3% of the cases. It produces solutions for moderately constrained systems (i.e., the deadline falls within 10% of the optimum makespan) in 88.1% of the cases.
• Efficient and Exact Design Space Exploration for Heterogeneous and Multi-Bus Platforms
  
  • Gharbi Amna
  • Enrici Andrea
  • Uscumlic Bogdan
  • Apvrille Ludovic
  • Pacalet Renaud
  , 2020. Design Space Exploration of data-flow Systems-on-Chip either focuses on classical shared bus or on complex network-on-chip (NoC) architectures. A lack of research work exists that targets segmented bus architectures. These offer performance improvements (latency, power consumption) with respect to a shared bus, while employing much simpler communication structures and algorithms than a NoC. Despite the lack in the research work, segmented buses are popular in multi-processor systems and in FPGA interconnects. This paper fills this lack with two contributions. First, we propose a Satisfiability Modulo Theory (SMT) formulation. Secondly, we provide a technique to reduce the design-space explosion problem that is portable to other formulations (e.g., ILP, MILP) and to problems where the scheduling on units (e.g., bus, CPU) is multiplexed in time. We integrated these contributions in a state-of-the-art design tool that we employ for evaluation purposes with a set of streaming applications and a MPSoC platform. The resulting framework can study the performance of fixed interconnects as well as determine the optimal architecture among a set of candidates. Our reduction technique improves considerably the scalability of DSE. For our testbench, we reduce the SMT solver run-time from 20 up to 589 times.
• A Multiclass Classification Approach to Label Ranking
  
  • Clémençon Stéphan
  • Vogel Robin
  , 2020, 108, pp.1421-1430. In multiclass classification, the goal is to learn how to predict a random label Y , valued in Y = {1, . . . , K} with K ≥ 3, based upon observing a r.v. X, taking its values in R q with q ≥ 1 say, by means of a classification rule g : R q → Y with minimum probability of error P{Y 6= g(X)}. However, in a wide variety of situations, the task targeted may be more ambitious, consisting in sorting all the possible label values y that may be assigned to X by decreasing order of the posterior probability ηy(X) = P{Y = y | X}. This article is devoted to the analysis of this statistical learning problem, halfway between multiclass classification and posterior probability estimation (regression) and referred to as label ranking here. We highlight the fact that it can be viewed as a specific variant of ranking median regression (RMR), where, rather than observing a random permutation Σ assigned to the input vector X and drawn from a BradleyTerry-Luce-Plackett model with conditional preference vector (η1(X), . . . , ηK(X)), the sole information available for training a label ranking rule is the label Y ranked on top, namely Σ−1 (1). Inspired by recent results in RMR, we prove that under appropriate noise conditions, the One-Versus-One (OVO) approach to multiclassification yields, as a by-product, an optimal ranking of the labels with overwhelming probability. Beyond theoretical guarantees, the relevance of the approach to label ranking promoted in this article is supported by experimental results.
• The Area of the Convex Hull of Sampled Curves: a Robust Functional Statistical Depth Measure
  
  • Staerman Guillaume
  • Mozharovskyi Pavlo
  • Clémençon Stéphan
  , 2020, PMLR 108:570-579. With the ubiquity of sensors in the IoT era, statistical observations are becoming increasingly available in the form of massive (multivariate) time-series. Formulated as unsupervised anomaly detection tasks, an abundance of applications like aviation safety management, the health monitoring of complex infrastructures or fraud detection can now rely on such functional data, acquired and stored with an ever finer granularity. The concept of \textit{statistical depth}, which reflects centrality of an arbitrary observation w.r.t. a statistical population may play a crucial role in this regard, anomalies corresponding to observations with ’small’ depth. Supported by sound theoretical and computational developments in the recent decades, it has proven to be extremely useful, in particular in functional spaces. However, most approaches documented in the literature consist in evaluating independently the centrality of each point forming the time series and consequently exhibit a certain insensitivity to possible shape changes.In this paper, we propose a novel notion of functional depth based on the area of the convex hull of sampled curves, capturing gradual departures from centrality, even beyond the envelope of the data, in a natural fashion.We discuss practical relevance of commonly imposed axioms on functional depths and investigate which of them are satisfied by the notion of depth we promote here. Estimation and computational issues are also adressed and various numerical experiments provide empirical evidence of the relevance of the approach proposed.
• Combining STPA with SysML Modeling
  
  • Rey de Souza Fellipe Guilherme
  • de Melo Bezerra Juliana
  • Hirata Celso Massaki
  • de Saqui-Sannes Pierre
  • Apvrille Ludovic
  , 2020. System-Theoretic Process Analysis (STPA) is a technique , based on System-Theoretic Accident Model and Process (STAMP), to identify hazardous control actions, loss scenarios, and safety requirements. STPA is considered a rather complex technique and lacks formalism, but there exists a growing interest in using STPA in certifications of safety-critical systems development. SysML is a modeling language for systems engineering. It enables representing models for analysis, design, verification, and validation of systems. In particular, the free software TTool and the model-checker UPPAAL enable formal verification of SysML models. This paper proposes a method that combines STPA and SysML modeling activities in order to allow simulation and formal verification of systems' models. An automatic door system serves as example to illustrate the effectiveness of the proposed approach.
• Nonlinear uncertainty propagation of on-wafer mixed-mode S parameter measurements using Multimode-TRL calibration
  
  • Pham Thi Dao
  • Allal Djamel
  • Ziade Francois
  • Bergeault Eric
  , 2020, pp.1-2. Full characterization of differential microwave circuits is traditionally performed by measuring their mixed mode scattering (S) parameters using a four-port vector network analyzer. This paper describes an evaluation of the uncertainty of on-wafer mixed-mode S parameter measurements carried out using a Multimode Thru-Reflect-Line (TRL) calibration kit based on coupled coplanar waveguide (CCPW) standards, developed on silicon dioxide substrate in the Ground-Signal-Ground-Signal-Ground (GSGSG) configuration and associated verification standards. In addition to the influence quantities linked with the measurement instrument and the uncertainty due to the calibration standards, measurement repeatability is evaluated. For the first time, potentially nonlinear uncertainty propagation through the Multimode-TRL calibration algorithm is shown, which can make it unsuitable when using embedded balanced calibration standards and degraded input uncertainties (10.1109/CPEM49742.2020.9191841)
  DOI : 10.1109/CPEM49742.2020.9191841
• Seamless Integration between Real-time Analyses and Systems Engineering with the PST Approach
  
  • Caron Françoise
  • Maxim Cristian
  • Blouin Dominique
  • Crisafulli Paolo
  IEEE SYSCON2020, IEEE, 2020. Cyber-physical systems are becoming more and more connected as it is the case for the railway domain. This introduces the need for smart monitoring and communication services besides the traditional control-command functions. As a consequence, more complex architectures are required to leverage connected technologies. This challenges traditional processes used to develop such systems. Model-based system engineering (MBSE) has been proposed to tackle this challenge by allowing virtual integration of the system in order to discover flaws earlier in the development process thus reducing rework costs. MBSE requires that several models of the system are conjointly used to assess architectures at different levels of abstraction starting from the overall system level down to specialty domains. However, there is currently a lack of approaches for using these models conjointly to perform virtual integration. In this paper, we present the PST approach allowing seamless integration between systems engineering and real-time specialty domain analyses to support virtual integration. The approach is illustrated with a system from the railway domain; the on-board equipment of the European Train Control System (ETCS).
• Learning to Cluster under Domain Shift
  
  • Menapace Willi
  • Lathuilière Stéphane
  • Ricci Elisa
  , 2020. While unsupervised domain adaptation methods based on deep architectures have achieved remarkable success in many computer vision tasks, they rely on a strong assumption, i.e. labeled source data must be available. In this work we overcome this assumption and we address the problem of transferring knowledge from a source to a target domain when both source and target data have no annotations. Inspired by recent works on deep clustering, our approach leverages information from data gathered from multiple source domains to build a domain-agnostic clustering model which is then refined at inference time when target data become available. Specifically, at training time we propose to optimize a novel information-theoretic loss which, coupled with domain-alignment layers, ensures that our model learns to correctly discover semantic labels while discarding domain-specific features. Importantly, our architecture design ensures that at inference time the resulting source model can be effectively adapted to the target domain without having access to source data, thanks to feature alignment and self-supervision. We evaluate the proposed approach in a variety of settings, considering several domain adaptation benchmarks and we show that our method is able to automatically discover relevant semantic information even in presence of few target samples and yields state-of-the-art results on multiple domain adaptation benchmarks.
• Online Continual Learning under Extreme Memory Constraints
  
  • Fini Enrico
  • Lathuilière Stéphane
  • Sangineto Enver
  • Nabi Moin
  • Ricci Elisa
  , 2020. Continual Learning (CL) aims to develop agents emulating the human ability to sequentially learn new tasks while being able to retain knowledge obtained from past experiences. In this paper, we introduce the novel problem of Memory-Constrained Online Continual Learning (MC-OCL) which imposes strict constraints on the memory overhead that a possible algorithm can use to avoid catastrophic forgetting. As most, if not all, previous CL methods violate these constraints, we propose an algorithmic solution to MC-OCL: Batch-level Distillation (BLD), a regularization-based CL approach, which effectively balances stability and plasticity in order to learn from data streams, while preserving the ability to solve old tasks through distillation. Our extensive experimental evaluation, conducted on three publicly available benchmarks, empirically demonstrates that our approach successfully addresses the MC-OCL problem and achieves comparable accuracy to prior distillation methods requiring higher memory overhead.
• TRADI: Tracking deep neural network weight distributions
  
  • Franchi Gianni
  • Bursuc Andrei
  • Aldea Emanuel
  • Dubuisson Séverine
  • Bloch Isabelle
  , 2020. During training, the weights of a Deep Neural Network (DNN) are optimized from a random initialization towards a nearly optimum value minimizing a loss function. Only this final state of the weights is typically kept for testing, while the wealth of information on the geometry of the weight space, accumulated over the descent towards the minimum is discarded. In this work we propose to make use of this knowledge and leverage it for computing the distributions of the weights of the DNN. This can be further used for estimating the epistemic uncertainty of the DNN by aggregating predictions from an ensemble of networks sampled from these distributions. To this end we introduce a method for tracking the trajectory of the weights during optimization, that does neither require any change in the architecture, nor in the training procedure. We evaluate our method, TRADI, on standard classification and regression benchmarks, and on out-of-distribution detection for classification and semantic segmentation. We achieve competitive results, while preserving computational efficiency in comparison to ensemble approaches. (10.48550/arXiv.1912.11316)
  DOI : 10.48550/arXiv.1912.11316
• 1.3-µm passively mode-locked quantum dot lasers epitaxially grown on silicon: gain properties and optical feedback stabilization
  
  • Dong Bozhang
  • de Labriolle Xavier C
  • Liu Songtao
  • Dumont Mario
  • Huang Heming
  • Duan Jianan
  • Norman Justin C
  • Bowers John E
  • Grillot Frédéric
  Journal of Physics: Photonics, IOP Science, 2020, 2. This work reports on an investigation of the optical feedback in an InAs/InGaAs passively mode-locked quantum dot (QD) laser epitaxially grown on silicon. Under the stably-resonant optical feedback condition, experiments demonstrate that the radio-frequency linewidth is narrowed whatever the bias voltage applied on the saturable absorber (SA) is; on the other hand, the effective linewidth enhancement factor of the device increases with the reverse bias voltage on the SA, hence it is observed that such an increase influences the mode-locking dynamic and the stability of device under optical feedback. This work gives insights for stabilizing epitaxial QD mode-locked lasers on silicon which is meaningful for their applications in future large-scale silicon electronic and photonic applications requiring low power consumption as well as for high-speed photonic analog-to-digital conversion, intrachip/interchip optical clock distribution and recovery. (10.1088/2515-7647/aba5a6)
  DOI : 10.1088/2515-7647/aba5a6
• A Cognitive Control System for Managing Runtime Uncertainty in Self-Integrating Autonomic Systems
  
  • Pol Marius
  • Diaconescu Ada
  , 2020.
• Quantum weak coin flipping with a single photon
  
  • Bozzio Mathieu
  • Chabaud Ulysse
  • Kerenidis Iordanis
  • Diamanti Eleni
  Physical Review A, American Physical Society, 2020, 102 (2), pp.022414. Weak coin flipping is among the fundamental cryptographic primitives which ensure the security of modern communication networks. It allows two mistrustful parties to remotely agree on a random bit when they favor opposite outcomes. Unlike other two-party computations, one can achieve information-theoretic security using quantum mechanics only: both parties are prevented from biasing the flip with probability higher than $1/2+\epsilon$, where $\epsilon$ is arbitrarily low. Classically, the dishonest party can always cheat with probability $1$ unless computational assumptions are used. Despite its importance, no physical implementation has been proposed for quantum weak coin flipping. Here, we present a practical protocol that requires a single photon and linear optics only. We show that it is fair and balanced even when threshold single-photon detectors are used, and reaches a bias as low as $\epsilon=1/\sqrt{2}-1/2\approx 0.207$. We further show that the protocol may display quantum advantage over a few hundred meters with state-of-the-art technology. (10.1103/PhysRevA.102.022414)
  DOI : 10.1103/PhysRevA.102.022414
• A Decentralized Approach to Explanatory Artificial Intelligence for Autonomic Systems
  
  • Houze Etienne
  • Diaconescu Ada
  • Dessalles Jean-Louis
  • Menga David
  • Schumann Mathieu
  , 2020, pp.115-120. (10.1109/ACSOS-C51401.2020.00041)
  DOI : 10.1109/ACSOS-C51401.2020.00041
• Online Trajectory Optimization Using Inexact Gradient Feedback for Time-Varying Environments
  
  • Nutalapati Mohan Krishna
  • Bedi Amrit Singh
  • Rajawat Ketan
  • Coupechoux Marceau
  IEEE Transactions on Signal Processing, Institute of Electrical and Electronics Engineers, 2020, 68, pp.4824-4838. This paper considers the problem of online trajectory design under time-varying environments. We formulate the general trajectory optimization problem within the framework of time-varying constrained convex optimization and propose a novel version of online gradient ascent algorithm for such problems. Moreover, the gradient feedback is noisy and allows to use the proposed algorithm for a range of practical applications where it is difficult to acquire the true gradient. Since we are interested in constrained online convex optimization, we carefully select the step size at each iteration so that the iterates stay feasible. In contrast to the most available literature, we present the offline sublinear regret of the proposed algorithm up to the path length variations of the optimal offline solution, the cumulative gradient, and the error in the gradient variations. Furthermore, we establish a lower bound on the offline dynamic regret, which defines the optimality of any trajectory. To show the efficacy of the proposed algorithm, we consider two practical problems of interest. First, we consider a device to device (D2D) communications setting, and the goal is to design a user trajectory while maximizing its connectivity to the internet. This problem is of vital interest, due to the surge in data-intensive applications in smartphones, and the consistent internet connectivity is becoming essential. For this problem, we consider a pair of pedestrians connected through a D2D link for data exchange applications such as file transfer, video calling, and online gaming, etc. The second problem is associated with the online planning of energy-efficient trajectories for unmanned surface vehicles (USV) under strong disturbances in ocean environments with both static and dynamic goal locations. We consider an unmanned surface vehicle (USV) operating in an ocean environment to traverse from start to destination. Different from the state of the art trajectory planning algorithms that entail planning and re-planning the full trajectory using the forecast data at each time instant, the proposed algorithm is entirely online and relies mostly on the current ocean velocity measurements at the vehicle locations. The detailed simulation results demonstrate the significance of the proposed algorithm on synthetic and real data sets. (10.1109/TSP.2020.3015276)
  DOI : 10.1109/TSP.2020.3015276
• Finite Open-world Query Answering with Number Restrictions
  
  • Amarilli Antoine
  • Benedikt Michael
  ACM Transactions on Computational Logic, Association for Computing Machinery, 2020, 21 (4), pp.1-73. (10.1145/3365834)
  DOI : 10.1145/3365834
• Connecting Knowledge Compilation Classes and Width Parameters
  
  • Amarilli Antoine
  • Capelli Florent
  • Monet Mikaël
  • Senellart Pierre
  Theory of Computing Systems, Springer Verlag, 2020. The field of knowledge compilation establishes the tractability of many tasks by studying how to compile them to Boolean circuit classes obeying some requirements such as structuredness, decomposability, and determinism. However, in other settings such as intensional query evaluation on databases, we obtain Boolean circuits that satisfy some width bounds, e.g., they have bounded treewidth or pathwidth. In this work, we give a systematic picture of many circuit classes considered in knowledge compilation and show how they can be systematically connected to width measures, through upper and lower bounds. Our upper bounds show that boundedtreewidth circuits can be constructively converted to d-SDNNFs, in time linear in the circuit size and singly exponential in the treewidth; and that bounded-pathwidth circuits can similarly be converted to uOBDDs. We show matching lower bounds on the compilation of monotone DNF or CNF formulas to structured targets, assuming a constant bound on the arity (size of clauses) and degree (number of occurrences of each variable): any d-SDNNF (resp., SDNNF) for such a DNF (resp., CNF) must be of exponential size in its treewidth, and the same holds for uOBDDs (resp., n-OBDDs) when considering pathwidth. Unlike most previous work, our bounds apply to any formula of this class, not just a well-chosen family. Hence, we show that pathwidth and treewidth respectively characterize the efficiency of compiling monotone DNFs to uOBDDs and d-SDNNFs with compilation being singly exponential in the corresponding width parameter. We also show that our lower bounds on CNFs extend to unstructured compilation targets, with an exponential lower bound in the treewidth (resp., pathwidth) when compiling monotone CNFs of constant arity and degree to DNNFs (resp., nFBDDs). (10.1007/s00224-019-09930-2)
  DOI : 10.1007/s00224-019-09930-2
• Challenge Codes for Physically Unclonable Functions with Gaussian Delays: A Maximum Entropy Problem
  
  • Schaub Alexander
  • Rioul Olivier
  • Danger Jean-Luc
  • Guilley Sylvain
  • Boutros Joseph J.
  Advances in Mathematics of Communications, AIMS, 2020, 14 (3), pp.491-505. In this paper, motivated by a security application on physically unclonable functions, we evaluate the distributions and Rényi entropies of signs of scalar products of i.i.d. Gaussian random variables against binary codewords 2 {±1} n. The exact distributions are determined for small values of n and upper bounds are provided by linking this problem to the study of Boolean threshold functions. Finally, Monte-Carlo simulations are used to approximate the distribution up to n = 10.
• SAR Image Despeckling by Deep Neural Networks: from a Pre-Trained Model to an End-to-End Training Strategy
  
  • Dalsasso Emanuele
  • Yang Xiangli
  • Denis Loïc
  • Tupin Florence
  • Yang Wen
  Remote Sensing, MDPI, 2020, 12 (16), pp.2636. (10.3390/rs12162636)
  DOI : 10.3390/rs12162636
• Design of an Integrated Platform for Mapping Residential Exposure to Rf-Emf Sources
  
  • Regrain Corentin
  • Caudeville Julien
  • de Seze René
  • Guedda Mohammed
  • Chobineh Amirreza
  • de Doncker Philippe
  • Petrillo Luca
  • Chiaramello Emma
  • Parazzini Marta
  • Joseph Wout
  • Aerts Sam
  • Huss Anke
  • Wiart Joe
  International Journal of Environmental Research and Public Health, MDPI, 2020, 17 (15), pp.5339. (10.3390/ijerph17155339)
  DOI : 10.3390/ijerph17155339
• Everlasting Secure Key Agreement with performance beyond QKD in a Quantum Computational Hybrid security model
  
  • Vyas Nilesh
  • Alleaume Romain
  , 2020. Extending the functionality and overcoming the performance limitation under which QKD can operate requires either quantum repeaters or new security models. Investigating the latter option, we introduce the \textit{Quantum Computational Hybrid} (QCH) security model, where we assume that computationally secure encryption may only be broken after time much longer than the coherence time of available quantum memories. We propose an explicit $d$-dimensional key distribution protocol, that we call MUB-\textit{Quantum Computational Timelock} (MUB-QCT) where one bit is encoded on a qudit state chosen among $d+1$ mutually unbiased bases (MUBs). Short-term-secure encryption is used to share the basis information with legitimate users while keeping it unknown from Eve until after her quantum memory decoheres. This allows reducing Eve's optimal attack to an immediate measurement followed by post-measurement decoding. \par We demonstrate that MUB-QCT enables everlasting secure key distribution with input states containing up to $O(\sqrt{d})$ photons. This leads to a series of important improvements when compared to QKD: on the functional side, the ability to operate securely between one sender and many receivers, whose implementation can moreover be untrusted; significant performance increase, characterized by a $O(\sqrt{d})$ multiplication of key rates and an extension by $25 {\rm} km \times \log(d)$ of the attainable distance over fiber. Implementable with a large number of modes with current or near-term multimode photonics technologies, the MUB-QCT construction has the potential to provide a radical shift to the performance and practicality of quantum key distribution.
• Misbehavior detection for cooperative intelligent transport systems (C-ITS)
  
  • Kamel Joseph
  , 2020. Cooperative Intelligent Transport Systems (C-ITS) is an upcoming technology that will change our driving experience in the near future. In such systems, vehicles cooperate by exchanging Vehicle-to-X communication (V2X) messages over the vehicular network. Safety applications use the data in these messages to detect and avoid dangerous situations on time. Therefore, it is crucial that the data in V2X messages is secure and accurate.In the current C-ITS system, the messages are signed with digital keys to ensure authenticity. However, authentication does not ensure the correctness of the data. A genuine vehicle could have a faulty sensor and therefore send inaccurate information. An attacker could also obtain legitimate keys by hacking into the on-board unit of his vehicle and therefore transmit signed malicious messages.Misbehavior Detection in C-ITS is an active research topic aimed at ensuring the correctness of the exchanged V2X messages. It consists of monitoring data semantics of the exchanged messages to detect and identify potential misbehaving entities. The detection process is divided into multiple steps. Local detection consists of first performing plausibility and consistency checks on the received V2X messages. The results of these checks are then fused using a local detection application. The application is able to identify various V2X anomalies. If an anomaly is detected, the vehicle will collect the needed evidence and create a misbehavior report. This report is then sent to a cloud based misbehavior authority.This authority has a goal of ensuring the correct operation of the C-ITS system and mitigating the effects of attacks. It will first collect the misbehavior reports from vehicles and would then investigate the event and decide on the suitable reaction.In this thesis, we evaluate and contribute to the local, reporting and global steps of the misbehavior detection process.
• A Survey on the Use of 2D Touch Interfaces for Musical Expression
  
  • Schwarz Diemo
  • Liu Wanyu
  • Bevilacqua Frédéric
  , 2020. Expressive 2D multi-touch interfaces have in recent years moved from research prototypes to industrial products, from repurposed generic computer input devices to controllers specially designed for musical expression. A host of practitioners use this type of devices in many different ways, with different gestures and sound synthesis or transformation methods. In order to get an overview of existing and desired usages, we launched an on-line survey that collected 37 answers from practitioners in and outside of academic and design communities. In the survey we inquired about the participants' devices, their strengths and weaknesses , the layout of control dimensions, the used gestures and mappings, the synthesis software or hardware, and the use of audio descriptors and machine learning. The results can inform the design of future interfaces, gesture analysis and mapping, and give directions for the need and use of machine learning for user adaptation.
• Why Facial Recognition Algorithms Can't be Perfectly Fair
  
  • Maxwell Winston
  • Clémençon Stéphan
  The Conversation France, The Conversation Media Group [•2015-....], 2020. In June 2020, a facial recognition algorithm led to the wrongful arrest of Robert Williams, an African-American, for a crime he did not commit. After a shoplifting incident in in a pricey area of Detroit, Michigan, his driver’s license photo was wrongly matched with a blurry video of the perpetrator. Police released him after several hours and apologised, but the episode raises serious questions about the accuracy of visual recognition algorithms. The troubling aspect of the story is that facial recognition algorithms have been shown to be less accurate for black faces than for white ones. But why do facial recognition algorithms make more mistakes for Blacks than whites, and what can be done about it?
• On Ensemble Techniques for Data Stream Regression
  
  • Gomes Heitor Murilo
  • Montiel Jacob
  • Mastelini Saulo Martiello
  • Pfahringer Bernhard
  • Bifet Albert
  , 2020, pp.1--8. An ensemble of learners tends to exceed the predictive performance of individual learners. This approach has been explored for both batch and online learning. Ensembles methods applied to data stream classification were thoroughly investigated over the years, while their regression counterparts received less attention in comparison. In this work, we discuss and analyze several techniques for generating, aggregating, and updating ensembles of regressors for evolving data streams. We investigate the impact of different strategies for inducing diversity into the ensemble by randomizing the input data (resampling, random subspaces and random patches). On top of that, we devote particular attention to techniques that adapt the ensemble model in response to concept drifts, including adaptive window approaches, fixed periodical resets and randomly determined windows. Extensive empirical experiments show that simple techniques can obtain similar predictive performance to sophisticated algorithms that rely on reactive adaptation (i.e., concept drift detection and recovery). (10.1109/IJCNN48605.2020.9206756)
  DOI : 10.1109/IJCNN48605.2020.9206756