Publications

2023

• Integration of Heterogeneous Components for Co-Simulation
  
  • Jerray Jawher
  • Ameur-Boulifa Rabea
  • Apvrille Ludovic
  , 2023, 1, pp.637-644. Because of their complexity, embedded systems are designed with sub-systems or components taken in charge by different development teams or entities and with different modeling frameworks and simulation tools, depending on the characteristics of each component. Unfortunately, this diversity of tools and semantics makes the integration of these heterogeneous components difficult. Thus, to evaluate their integration before their hardware or software is available, one solution would be to merge them into a common modeling framework. Yet, such a holistic environment supporting many computation and computation semantics seems hard to settle. Another solution we investigate in this paper is to generically link their respective simulation environments in order to keep the strength and semantics of each component environment. The paper presents a method to simulate heterogeneous components of embedded systems in real-time. These components can be described at any abstraction level. Our main contribution is a generic glue that can analyze in real-time the state of different simulation environments and accordingly enforce the correct communication semantics between components. (10.5220/0012134800003538)
  DOI : 10.5220/0012134800003538
• How About Kind of Generating Hedges using End-to-End Neural Models?
  
  • Abulimiti Alafate
  • Clavel Chloé
  • Cassell Justine
  , 2023. Hedging is a strategy for softening the impact of a statement in conversation. In reducing the strength of an expression, it may help to avoid embarrassment (more technically, ``face threat'') to one's listener. For this reason, it is often found in contexts of instruction, such as tutoring. In this work, we develop a model of hedge generation based on i) fine-tuning state-of-the-art language models trained on human-human tutoring data, followed by ii) reranking to select the candidate that best matches the expected hedging strategy within a candidate pool using a hedge classifier. We apply this method to a natural peer-tutoring corpus containing a significant number of disfluencies, repetitions, and repairs. The results show that generation in this noisy environment is feasible with reranking. By conducting an error analysis for both approaches, we reveal the challenges faced by systems attempting to accomplish both social and task-oriented goals in conversation.
• Un cadre flexible pour l'apprentissage automatique interprétable : application à la classification d'images et d'audio
  
  • Parekh Jayneel
  , 2023. Les systèmes d'apprentissage automatique, et en particulier les réseaux de neurones, ont rapidement développé leur capacité à résoudre des problèmes d'apprentissage complexes. Par conséquent, ils sont intégrés dans la société avec une influence de plus en plus grande sur tous les niveaux de l'expérience humaine. Cela a entraîné la nécessité d'acquérir des informations compréhensibles par l'homme dans leur processus de prise de décision pour s'assurer que les décisions soient prises de manière éthique et fiable. L'étude et le développement de méthodes capables de générer de telles informations constituent de manière générale le domaine de l'apprentissage automatique interprétable.Cette thèse vise à développer un nouveau cadre pour aborder deux problématiques majeures dans ce domaine, l'interprétabilité post-hoc et par conception. L'interprétabilité post-hoc conçoit des méthodes pour analyser les décisions d'un modèle prédictif pré-entraîné, tandis que l'interprétabilité par conception vise à apprendre un modèle unique capable à la fois de prédiction et d'interprétation. Pour ce faire, nous étendons la formulation traditionnelle de l'apprentissage supervisé pour inclure l'interprétation en tant que tâche supplémentaire en plus de la prédiction, chacune étant traitée par des modèles distincts, mais liés, un prédicteur et un interpréteur. Fondamentalement, l'interpréteur dépend du prédicteur à travers ses couches cachées et utilise un dictionnaire de concepts comme représentation pour l'interprétation avec la capacité de générer des interprétations locales et globales.Le cadre est instancié séparément pour résoudre les problèmes d'interprétation dans le contexte de la classification d'images et de sons. Les deux systèmes ont fait l'objet d'une évaluation approfondie de leurs interprétations sur de multiples ensembles de données publics. Dans les deux cas, nous démontrons des performances de prédiction élevées, ainsi qu'une haute fidélité des interprétations. Bien qu'ils adhèrent à la même structure sous-jacente, les deux systèmes sont distinctement conçus pour l'interprétation. Le système d'interprétabilité des images fait avancer le protocole de découverte des concepts appris pour une meilleure compréhension, laquelle est évaluée qualitativement. De plus, il inclut un nouveau critère pour rendre les interprétations plus concises. Le système d'interprétabilité audio est, quant à lui, conçu avec une nouvelle représentation basée sur une factorisation matricielle non-négative pour faciliter les interprétations écoutables, tout en modélisant les objets audio composant une scène.
• A Deep Learning Approach for SAR Tomographic Imaging of Forested Areas
  
  • Berenger Zoé
  • Denis Loïc
  • Tupin Florence
  • Ferro-Famil Laurent
  • Huang Yue
  IEEE Geoscience and Remote Sensing Letters, IEEE - Institute of Electrical and Electronics Engineers, 2023, 20, pp.4007405. Synthetic aperture radar tomographic imaging reconstructs the three-dimensional reflectivity of a scene from a set of coherent acquisitions performed in an interferometric configuration. In forest areas, a high number of elements backscatter the radar signal within each resolution cell. To reconstruct the vertical reflectivity profile, state-of-the-art techniques perform a regularized inversion implemented in the form of iterative minimization algorithms. We show that lightweight neural networks can be trained to perform this inversion with a single feedforward pass, leading to fast reconstructions that could better scale to the amount of data provided by the future BIOMASS mission. We train our encoder-decoder network using simulated data and validate our technique on real L-band and P-band data. (10.1109/LGRS.2023.3293470)
  DOI : 10.1109/LGRS.2023.3293470
• Improved frequency comb operation of an InAs/GaAs hybrid multisection quantum dot laser on silicon
  
  • Renaud Thibaut
  • Huang Heming
  • Kurczveil G
  • Liang D
  • Beausoleil R G
  • Grillot Frédéric
  Applied Physics Letters, American Institute of Physics, 2023, 123 (1), pp.011105:1-011105:6. This work reports on a systematic investigation of the frequency comb enhancement in hybrid InAs/GaAs multisection quantum dot lasers on silicon. The colliding configuration provides an operating frequency at twice the fundamental frequency of the free-spectral range of the cold cavity. In particular, the contribution of the linewidth enhancement factor, or αH-factor, on the comb formation is investigated with respect to the reverse voltage and temperature conditions. When those parameters are varied, the formation of the combs is found to increase with respect to αH. In addition, we also demonstrate that this quantum dot laser exhibits a comb behavior, while the beatnote locking is not fully achieved. This effect is essentially due to the dispersion which is not fully compensated from the optical nonlinearities. These results bring further insights on comb and pulse formations in multisection quantum dot lasers, which is important for designing future light sources for on-chip and chip-to-chip optical interconnects. (10.1063/5.0143570)
  DOI : 10.1063/5.0143570
• Pruning and compression of multi-view content for immersive video coding
  
  • Milovanovic Marta
  , 2023. This thesis addresses the problem of efficient compression of immersive video content, represented with Multiview Video plus Depth (MVD) format. The Moving Picture Experts Group (MPEG) standard for the transmission of MVD data is called MPEG Immersive Video (MIV), which utilizes 2D video codecs to compress the source texture and depth information. Compared to traditional video coding, immersive video coding is more complex and constrained not only by trade-off between bitrate and quality, but also by the pixel rate. Because of that, MIV uses pruning to reduce the pixel rate and inter-view correlations and creates a mosaic of image pieces (patches). Decoder-side depth estimation (DSDE) has emerged as an alternative approach to improve the immersive video system by avoiding the transmission of depth maps and moving the depth estimation process to the decoder side. DSDE has been studied for the case of numerous fully transmitted views (without pruning). In this thesis, we demonstrate possible advances in immersive video coding, emphasized on pruning the input content. We go beyond DSDE and examine the distinct effect of patch-level depth restoration at the decoder side. We propose two approaches to incorporate decoder-side depth estimation (DSDE) on content pruned with MIV. The first approach excludes a subset of depth maps from the transmission, and the second approach uses the quality of depth patches estimated at the encoder side to distinguish between those that need to be transmitted and those that can be recovered at the decoder side. Our experiments show 4.63 BD-rate gain for Y-PSNR on average. Furthermore, we also explore the use of neural image-based rendering (IBR) techniques to enhance the quality of novel view synthesis and show that neural synthesis itself provides the information needed to prune the content. Our results show a good trade-off between pixel rate and synthesis quality, achieving the view synthesis improvements of 3.6 dB on average.
• Performance evaluation and resource allocation in millimeter waves device-to-device networks with beamforming
  
  • Quan Yibo
  , 2023. Device-to-Device (D2D) communication is a key technology for future wireless networks, allowingdevices to communicate directly without relying on a cellular infrastructure. Millimeter wave (mm-Wave) communication utilizes high-frequency radio, providing very large bandwidths for fast and reliable D2D transmissions. However, mmWave frequencies have high attenuation, requiring devices to have multiple antennas and perform beamforming. The success of beamforming requires beam training. The beam misalignment can impact the performance of the network. To address these challenges, our study focuses on the theoretical analysis of the performance of mmWave D2D communications within the context of Ultra-Reliable Low Latency Communications(URLLC). We use stochastic geometry and queuing theory to evaluate both spatial and temporalvariations in performance from two different perspectives: the instantaneous average properties of the random network and the global stability properties of a dynamic network with random service requests. For the dynamic properties, we focus on the stability condition of D2D network by introducing directional antennas arrays for the D2D users. The network is modeled based on a spatial birth-death process. For the instantaneous properties, we mainly care about the meta-distribution of the network, which is a metric that accounts for the spatial distribution of coverageprobability. We derive the meta-distribution of the effective rate as a statistical latency guarantee for URLLC communications, by considering both the training overhead and misalignment for a D2D network with beamforming. At last, we propose methods to choose the optimal number of antennas and to allocate resource for beam training.
• Here comes SAID: A SOME/IP Attention-based mechanism for Intrusion Detection
  
  • Alkhatib Natasha
  • Mushtaq Maria
  • Ghauch Hadi
  • Danger Jean-Luc
  , 2023, pp.462-467. The increasing connectivity among vehicles along with their rising complexity increases their attack surface and challenges their security. In this paper, we consider the problem of intrusion detection for SOME/IP protocol and present “SAID” a novel technique for the detection of anomalies from a large sequence of exchanged SOME/IP network packets. The proposed detector leverages a self-attention-based neural network to model the contextual dependencies between SOME/IP packets. For this purpose, we evaluate our proposed approach, by generating a simulated and manually annotated SOME/IP dataset, with several categories of attacks. The results of the extensive experiments indicate that our technique detects (with high accuracy) the majority of SOME/IP’s protocol violations, e.g., with an area-under-the-curve ≈0.8, and inference time ≈0.3 ms. A comparative study, including various state-of-the-art benchmark algorithms, shows that SAID shows better performance in detecting intrusions and enables parallelization. Our source code and data are available at: https://github.com/Alkhatibnatasha/supervised_detection_some_ip (10.1109/ICUFN57995.2023.10200508)
  DOI : 10.1109/ICUFN57995.2023.10200508
• Comparison of Data Cleansing Methods for Network DDoS Attacks Mitigation
  
  • Jamal Adonis
  • El Attar Ali
  • Chbib Fadlallah
  • Khatoun Rida
  , 2023, pp.459-464. A Distributed Denial of Service (DDoS) attack is a malicious attempt to disrupt the normal traffic of a targeted server, service, or network by overwhelming it with a flood of requests from multiple compromised internet-connected devices, such as distributed servers, personal computers, and Internet of Things devices. One of the methods used to defend against DDoS attacks is traffic redirection to a Scrubbing Center (SC) for further inspection and mitigation. In this research, we present a novel scrubbing method that employs machine learning models to detect DDoS attacks. We propose using three machine learning algorithms, Random Forest, Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost), and combine them with three feature selection techniques, Analysis of Variance (ANOVA), Principal Component Analysis (PCA), and Kendall's Rank Correlation. Our results indicate that a combination of Kendall's Rank Correlation as a feature selector with SVM, XGBoost, and Random Forest models achieved a high F1 score. (10.1109/CoDIT58514.2023.10284093)
  DOI : 10.1109/CoDIT58514.2023.10284093
• General Knowledge Representation and Sharing, with Illustrations in Risk/Emergency Management
  
  • Martin Philippe A
  • Tanzi Tullio
  Sustainability, MDPI, 2023, 15 (14). Many decision-making tasks, including the sustainability-oriented ones and those related to the management of risks or emergencies, must gather, integrate, and analyze an important amount of information of various kinds and origins. Hence, how should information be best organized and shared by agents – people or software – for all and only the pieces of information looked for by these agents to maximize their retrieval, reuse, organization and analysis by these agents? To that end, various logic-based knowledge representation (KR) and sharing (KS) techniques, and hence KR bases, have been used. However, most KS researchers focus on what this article defines as “restricted KR and KS”, where information providers and consumers can or have to discuss for solving information ambiguities and other problems. The first part of this article highlights the usefulness of “general KR and KS” and, for supporting them, provides a panorama of complementary techniques, and hence, indirectly, best practices or kinds of tools to use for general KS purposes. These techniques collectively answer research questions about how to support Web users in the collaborative building of KR bases. The second part uses the risk/emergency management domain to illustrate the ways different types of information can be represented to support general KS. (10.3390/su151410803)
  DOI : 10.3390/su151410803
• Probabilistic Shaping over Multi-Dimensional Constellations for Optical Fiber Transmissions: Trade-offs and Insights
  
  • Liu Jingtian
  • Awwad Elie
  • Jaouën Yves
  , 2023. In this work, we combine probabilistic constellation shaping (PCS) using a Maxwell Boltzmann distribution with multi-dimensional modulations with the purpose of increasing performance gains in both the linear and non-linear regimes of optical fiber transmission systems. We particularly study the case of polarization division multiplexed long-haul coherent systems. By applying set partitioning and energy constraints on four-dimensional (4D) modulations followed by probabilistic shaping, we give insights into constructing novel modulations with enhanced performance compared to conventional PCS constellations in which shaping is performed separately over each real dimension.
• Relaxing dispersion pre-distorsion constraints of receiver-based power profile estimators
  
  • Tomczyk Louis
  • Awwad Élie
  • Ramantanis Pétros
  • Ware Cédric
  , 2023, pp.1-6. In this work, we offer insights on the advantage of using dispersion pre-distortion (DPD) in correlation-based power profile estimation. We also introduce a method to avoid the use of DPD by replacing it with a receiver-side digital correction of the estimated profile. We assess the accuracy of power anomaly estimation and location of the two methods, namely with or without pre-dipsersion at the transmitter side. Our method demonstrates that both approaches exhibit a similar range of applicability by giving at most relative errors of loss estimation around 20% and location around 2% compared to the ground truth. (10.1109/OECC56963.2023.10209744)
  DOI : 10.1109/OECC56963.2023.10209744
• AIoT-based Neural Decoding and Neurofeedback for Accelerated Cognitive Training: Vision, Directions and Preliminary Results
  
  • Nguyen Van-Tam
  • Tartaglione Enzo
  • Dinh Tuan
  , 2023, pp.705-709. Attention and working memory, which are two fundamental components of cognitive basis, can be improved through cognitive training. In addition, thanks to neuroplasticity, neurons are able to adapt quickly to the demands placed on them. By developing new neural networks and strengthening important connections, a cognitive training program can measurably and permanently improve brain activity. In this paper, we present a concept of AIoT based neural decoding and neurofeedback to accelerate cognitive training, the preliminary results and research directions. The proposed concept is to design adequate tiny machine learning to extract the relevant features and characteristics from physiological signals. A tiny ML performs classification or recognition of relevant patterns, based on which the neurofeedback system is appropriately designed for more effective cognitive training (10.1109/SSP53291.2023.10208067)
  DOI : 10.1109/SSP53291.2023.10208067
• Optimized preprocessing and Tiny ML for Attention State Classification
  
  • Wang Yinghao
  • Nahon Rémi
  • Tartaglione Enzo
  • Mozharovskyi Pavlo
  • Nguyen Van-Tam
  , 2023, pp.695-699. Electroencephalography has been widely used to study mental processes such as attention, perception, and emotion. This is because mental state classification has important applications in many fields, including healthcare, human-computer interaction, and education.In this paper, we present a new approach to mental state classification from EEG signals by combining signal processing techniques and machine learning (ML) algorithms. We evaluate the performance of the proposed method on a dataset of EEG recordings collected during a cognitive load task. The results show that the proposed method achieves high accuracy in classifying mental states and outperforms state-of-the-art methods in terms of classification accuracy and computational efficiency (10.1109/SSP53291.2023.10207930)
  DOI : 10.1109/SSP53291.2023.10207930
• Successive Quantization of the Neural Network Equalizers in Optical Fiber Communication
  
  • Darweesh Jamal
  • Costa Nelson
  • Jaouën Yves
  • Napoli Antonio
  • Pedro Jaoa
  • Spinnler Bernhard
  • Yousefi Mansoor
  , 2023. pragmatic successive quantization approach is applied to a neural network equalizer in a 16-QAM dualpolarization fiber transmission experiment over a 9x50km TWC fiber link. Quantization at 5 bits reduces the complexity by 85%, with a negligible Q-factor penalty
• Infomathic
  
  • Zayana Karim
  • Queruel Régis
  • Michalak Pierre
  Quadrature, EDP Sciences, 2023. Depuis qu’il existe, l’outil informatique a souvent épaulé les mathématiciens, qu’il s’agisse d’implémenter une méthode d’approximation (calcul numérique d’une racine, d’une intégrale,. . .) ou de simuler un phénomène (de nature géométrique, probabiliste,. . .) pour vérifier ou établir une conjecture. Mais, et c’est un autre point sur lequel nous concentrerons ici notre attention, l’informatique aura également servi la cause des mathématiques en inspirant certains raisonnements ou en prenant à sa charge les pans entiers d’une démonstration. Nous allons illustrer ce fructueux partenariat par deux exemples accessibles dès les classes de lycée.
• Quantum Metrology Using Time-Frequency as Quantum Continuous Variables: Resources, Sub-Shot-Noise Precision and Phase Space Representation
  
  • Descamps Eloi
  • Fabre Nicolas
  • Keller Arne
  • Milman Pérola
  Physical Review Letters, American Physical Society, 2023, 131 (3), pp.030801. We study the role of the electromagnetic field’s frequency on the precision limits of time measurements from a quantum perspective, using single photons as a paradigmatic system. We demonstrate that a quantum enhancement of precision is possible only when combining both intensity and spectral resources and, in particular, that spectral correlations enable a quadratic scaling of precision with the number of probes. We identify the general mathematical structure of nonphysical states that achieve the Heisenberg limit and show how a finite spectral variance may cause a quantum-to-classical-like transition in precision scaling for pure states similar to the one observed for noisy systems. Finally, we provide a clear and consistent geometrical time-frequency phase space interpretation of our results, identifying what should be considered as spectral classical resources. (10.1103/PhysRevLett.131.030801)
  DOI : 10.1103/PhysRevLett.131.030801
• Multi-hop network with multiple decision centers under expected-rate constraints
  
  • Hamad Mustapha
  • Wigger Michèle
  • Sarkiss Mireille
  IEEE Transactions on Information Theory, Institute of Electrical and Electronics Engineers, 2023, 69 (7), pp.4255-4283. We consider a multi-hop distributed hypothesis testing problem with multiple decision centers (DCs) for testing against independence and where the observations obey some Markov chain. For this system, we characterize the fundamental type-II error exponents region, i.e., the type-II error exponents that the various DCs can achieve simultaneously, under expected-rate constraints. Our results show that this fundamental exponents region is boosted compared to the region under maximum-rate constraints, and that it depends on the permissible type-I error probabilities. When all DCs have equal permissible type-I error probabilities, the exponents region is rectangular and all DCs can simultaneously achieve their optimal type-II error exponents. When the DCs have different permissible type-I error probabilities, a tradeoff between the type-II error exponents at the different DCs arises. New achievability and converse proofs are presented. For the achievability, a new multiplexing and rate-sharing strategy is proposed. The converse proof is based on applying different change of measure arguments in parallel and on proving asymptotic Markov chains. For the special cases K∈{2,3} , and for arbitrary K≥2 when all permissible type-I error probabilities at the various DCs are equal, we provide simplified expressions for the exponents region; a similar simplification is conjectured for the general case (10.1109/TIT.2023.3238339)
  DOI : 10.1109/TIT.2023.3238339
• Wind power predictions from nowcasts to 4-hour forecasts: a learning approach with variable selection
  
  • Bouche Dimitri
  • Flamary Rémi
  • D’alché-Buc Florence
  • Plougonven Riwal
  • Clausel Marianne
  • Badosa Jordi
  • Drobinski Philippe
  Renewable Energy, Elsevier, 2023, 211, pp.938-947. We study short-term prediction of wind speed and wind power (every 10 minutes up to 4 hours ahead). Accurate forecasts for these quantities are crucial to mitigate the negative effects of wind farms' intermittent production on energy systems and markets. We use machine learning to combine outputs from numerical weather prediction models with local observations. The former provide valuable information on higher scales dynamics while the latter gives the model fresher and location-specific data. So as to make the results usable for practitioners, we focus on well-known methods which can handle a high volume of data. We study first variable selection using both a linear technique and a nonlinear one. Then we exploit these results to forecast wind speed and wind power still with an emphasis on linear models versus nonlinear ones. For the wind power prediction, we also compare the indirect approach (wind speed predictions passed through a power curve) and the indirect one (directly predict wind power). (10.1016/j.renene.2023.05.005)
  DOI : 10.1016/j.renene.2023.05.005
• Audio Signal Processing in the 21st Century
  
  • Richard Gaël
  • Smaragdis Paris
  • Gannot Sharon
  • Naylor Patrick A
  • Makino Shoji
  • Kellermann Walter
  • Sugiyama Akihiko
  IEEE Signal Processing Magazine, Institute of Electrical and Electronics Engineers, 2023. (10.1109/MSP.2023.3276171)
  DOI : 10.1109/MSP.2023.3276171
• Measuring Lexico-Semantic Alignment in Debates with Contextualized Word Representations
  
  • Garí Soler Aina
  • Labeau Matthieu
  • Clavel Chloé
  , 2023, pp.50-63. Dialog participants sometimes align their linguistic styles, e.g., they use the same words and syntactic constructions as their interlocutors. We propose to investigate the notion of lexico-semantic alignment: to what extent do speakers convey the same meaning when they use the same words? We design measures of lexico-semantic alignment relying on contextualized word representations. We show that they reflect interesting semantic differences between the two sides of a debate and that they can assist in the task of debate’s winner prediction. (10.18653/v1/2023.sicon-1.6)
  DOI : 10.18653/v1/2023.sicon-1.6
• Aesthetics in digital photography
  
  • Maitre Henri
  , 2023. Automatically evaluating the aesthetic qualities of a photograph is a current challenge for artificial intelligence technologies, yet it is also an opportunity to open up new economic and social possibilities. Aesthetics in Digital Photography presents theories developed over the last 25 centuries by philosophers and art critics, who have sometimes been governed by the objectivity of perception, and other times, of course, by the subjectivity of human judgement. It explores the advances that have been made in neuro-aesthetics and their current limitations. In the field of photography, this book puts aesthetic hypotheses up against experimental verification, and then critically examines attempts to "scientifically" measure this beauty. Special attention is paid to artificial intelligence techniques, taking advantage of machine learning methods and large databases. (10.1002/9781394225972)
  DOI : 10.1002/9781394225972
• A grounded theory of Community Package Maintenance Organizations
  
  • Zimmermann Théo
  • Falleri Jean-Rémy
  Empirical Software Engineering, Springer Verlag, 2023, 28 (4), pp.101. In many programming language ecosystems, developers rely more and more on external open source dependencies, made available through package managers. Key ecosystem packages that go unmaintained create a health risk for the projects that depend on them and for the ecosystem as a whole. Therefore, community initiatives can emerge to alleviate the problem by adopting packages in need of maintenance. The goal of our study is to explore such community initiatives, that we will designate from now on as Community Package Maintenance Organizations (CPMOs) and to build a theory of how and why they emerge, how they function and their impact on the surrounding ecosystems. To achieve this, we use a qualitative methodology called Grounded Theory. We have applied this methodology in two steps. First, on ``extant'' documents (documentation, discussions on public forums) originating from several CPMOs. From this data, we have built a theory of CPMOs, which we have then refined through interviews and reliability checks with CPMO participants. Our theory can inform developers willing to launch a CPMO in their own ecosystem and help current CPMO participants to better understand the state of the practice and what they could do better. It is a basis on which future research can be done on how to help open source ecosystems improve the maintenance status of their most important packages. (10.1007/s10664-023-10337-4)
  DOI : 10.1007/s10664-023-10337-4
• Online machine learning-based predictive maintenance for the railway industry
  
  • Le Nguyen Minh Huong
  , 2023. Being an effective long-distance mass transit, the railway will continue to flourish for its limited carbon footprint in the environment. Ensuring the equipment's reliability and passenger safety brings forth the need for efficient maintenance. Apart from the prevalence of corrective and periodic maintenance, predictive maintenance has come into prominence lately. Recent advances in machine learning and the abundance of data drive practitioners to data-driven predictive maintenance. The common practice is to collect data to train a machine learning model, then deploy the model for production and keep it unchanged afterward. We argue that such practice is suboptimal on a data stream. The unboundedness of the stream makes the model prone to incomplete learning. Dynamic changes on the stream introduce novel concepts unseen by the model and decrease its accuracy. The velocity of the stream makes manual labeling infeasible and disables supervised learning algorithms. Therefore, switching from a static, offline learning paradigm to an adaptive, online one is necessary, especially when new generations of connected trains continuously generating sensor data have already been a reality. We investigate the applicability of online machine learning for predictive maintenance on typical complex systems in the railway. First, we develop InterCE as an active learning-based framework that extracts cycles from an unlabeled stream by interacting with a human expert. Then, we implement a long short-term memory autoencoder to transform the extracted cycles into feature vectors that are more compact yet remain representative. Finally, we design CheMoc as a framework that continuously monitors the condition of the systems using online adaptive clustering. Our methods are evaluated on the passenger access systems on two fleets of passenger trains managed by the national railway company SNCF of France.
• Linewidth narrowing in self-injection-locked on-chip lasers
  
  • Alkhazraji Emad
  • Chow Weng W
  • Grillot Frédéric
  • Bowers John E
  • Wan Yating
  Light: Science and Applications, Nature Publishing Group, 2023, 12 (162), pp.1-10. Stable laser emission with narrow linewidth is of critical importance in many applications, including coherent communications, LIDAR, and remote sensing. In this work, the physics underlying spectral narrowing of self-injection-locked on-chip lasers to Hz-level lasing linewidth is investigated using a composite-cavity structure. Heterogeneously integrated III–V/SiN lasers operating with quantum-dot and quantum-well active regions are analyzed with a focus on the effects of carrier quantum confinement. The intrinsic differences are associated with gain saturation and carrier-induced refractive index, which are directly connected with 0- and 2-dimensional carrier densities of states. Results from parametric studies are presented for tradeoffs involved with tailoring the linewidth, output power, and injection current for different device configurations. Though both quantum-well and quantum-dot devices show similar linewidth-narrowing capabilities, the former emits at a higher optical power in the self-injection-locked state, while the latter is more energy-efficient. Lastly, a multi-objective optimization analysis is provided to optimize the operation and design parameters. For the quantum-well laser, minimizing the number of quantum-well layers is found to decrease the threshold current without significantly reducing the output power. For the quantum-dot laser, increasing the quantum-dot layers or density in each layer increases the output power without significantly increasing the threshold current. These findings serve to guide more detailed parametric studies to produce timely results for engineering design. (10.1038/s41377-023-01172-9)
  DOI : 10.1038/s41377-023-01172-9