Publications

2022

• Questioning the Validity of Summarization Datasets and Improving Their Factual Consistency
  
  • Guo Yanzhu
  • Clavel Chloé
  • Kamal Eddine Moussa
  • Vazirgiannis Michalis
  , 2022, pp.5716-5727. The topic of summarization evaluation has recently attracted a surge of attention due to the rapid development of abstractive summarization systems. However, the formulation of the task is rather ambiguous, neither the linguistic nor the natural language processing communities have succeeded in giving a mutually agreed-upon definition. Due to this lack of well-defined formulation, a large number of popular abstractive summarization datasets are constructed in a manner that neither guarantees validity nor meets one of the most essential criteria of summarization: factual consistency. In this paper, we address this issue by combining state-of-the-art factual consistency models to identify the problematic instances present in popular summarization datasets. We release SummFC, a filtered summarization dataset with improved factual consistency, and demonstrate that models trained on this dataset achieve improved performance in nearly all quality aspects. We argue that our dataset should become a valid benchmark for developing and evaluating summarization systems. (10.18653/v1/2022.emnlp-main.386)
  DOI : 10.18653/v1/2022.emnlp-main.386
• LogiTorch: A PyTorch-based library for logical reasoning on natural language
  
  • Helwe Chadi
  • Clavel Chloé
  • Suchanek Fabian
  , 2022. Logical reasoning on natural language is one of the most challenging tasks for deep learning models. There has been an increasing interest in developing new benchmarks to evaluate the reasoning capabilities of language models such as BERT. In parallel, new models based on transformers have emerged to achieve ever better performance on these datasets. However, there is currently no library for logical reasoning that includes such benchmarks and models. This paper introduces LogiTorch, a PyTorch-based library that includes different logical reasoning benchmarks, different models, as well as utility functions such as co-reference resolution. This makes it easy to directly use the preprocessed datasets, to run the models, or to finetune them with different hyperparameters. LogiTorch is open source and can be found on GitHub .
• Multi-Client Functional Encryption with Fine-Grained Access Control
  
  • Nguyen Ky
  • Phan Duong Hieu
  • Pointcheval David
  , 2022. Multi-Client Functional Encryption (MCFE) and Multi-Input Functional Encryption (MIFE) are very interesting extensions of Functional Encryption for practical purpose. They allow to compute joint function over data from multiple parties. Both primitives are aimed at applications in multiuser settings where decryption can be correctly output for users with appropriate functional decryption keys only. While the definitions for a single user or multiple users were quite general and can be realized for general classes of functions as expressive as Turing machines or all circuits, efficient schemes have been proposed so far for concrete classes of functions: either only for access control, i.e. the identity function under some conditions, or linear/quadratic functions under no condition. In this paper, we target classes of functions that explicitly combine some evaluation functions independent of the decrypting user under the condition of some access control. More precisely, we introduce a framework for MCFE with fine-grained access control and propose constructions for both single-client and multi-client settings, for inner-product evaluation and access control via Linear Secret Sharing Schemes (LSSS), with selective and adaptive security. The only known work that combines functional encryption in multiuser setting with access control was proposed by Abdalla et al. (Asiacrypt '20), which relies on a generic transformation from the single-client schemes to obtain MIFE schemes that suffer a quadratic factor of n (where n denotes the number of clients) in the ciphertext size. We follow a different path, via MCFE: we present a duplicate-and-compress technique to transform the single-client scheme and obtain a MCFE with fine-grained access control scheme with only a linear factor of n in the ciphertext size. Our final scheme thus outperforms the Abdalla et al.'s scheme by a factor n, as one can obtain MIFE from MCFE by making all the labels in MCFE a fixed public constant. The concrete constructions are secure under the SXDH assumption, in the random oracle model for the MCFE scheme, but in the standard model for the MIFE improvement. (10.1007/978-3-031-22963-3_4)
  DOI : 10.1007/978-3-031-22963-3_4
• STREAMRHF: Tree-Based Unsupervised Anomaly Detection for Data Streams
  
  • Nesic Stefan
  • Putina Andrian
  • Bahri Maroua
  • Huet Alexis
  • Manuel Jose
  • Rossi Dario
  • Sozio Mauro
  , 2022. We present STREAMRHF, an unsupervised anomaly detection algorithm for data streams. Our algorithm builds on some of the ideas of Random Histogram Forest (RHF), a state-of-the-art algorithm for batch unsupervised anomaly detection. STREAMRHF constructs a forest of decision trees, where feature splits are determined according to the kurtosis score of every feature. It irrevocably assigns an anomaly score to data points, as soon as they arrive, by means of an incremental computation of its random trees and the kurtosis scores of the features. This allows efficient online scoring and concept drift detection altogether. Our approach is tree-based which boasts several appealing properties, such as explainability of the results. We conduct an extensive experimental evaluation on multiple datasets from different real-world applications. Our evaluation shows that our streaming algorithm achieves comparable average precision to RHF while outperforming state-of-the-art streaming approaches for unsupervised anomaly detection with furthermore limited computational complexity.
• CAN-BERT do it? Controller Area Network Intrusion Detection System based on BERT Language Model
  
  • Alkhatib Natasha
  • Mushtaq Maria
  • Ghauch Hadi
  • Danger Jean-Luc
  , 2022, pp.1-8. Due to the rising number of sophisticated customer functionalities, electronic control units (ECUs) are increasingly integrated into modern automotive systems. However, the high connectivity between the in-vehicle and the external networks paves the way for hackers who could exploit in-vehicle network protocols' vulnerabilities. Among these protocols, the Controller Area Network (CAN), known as the most widely used in-vehicle networking technology, lacks encryption and authentication mechanisms, making the communications delivered by distributed ECUs insecure. Inspired by the outstanding performance of bidirectional encoder representations from transformers (BERT) for improving many natural language processing tasks, we propose in this paper “CAN-BERT”, a deep learning based network intrusion detection system, to detect cyber attacks on CAN bus protocol. We show that the BERT model can learn the sequence of arbitration identifiers (IDs) in the CAN bus for anomaly detection using the “masked language model” unsupervised training objective. The experimental results on the “Car Hacking: Attack & Defense Challenge 2020” dataset show that “CAN-BERT” outperforms state-of-the-art approaches. In addition to being able to identify in-vehicle intrusions in real-time within 0.8 ms to 3 ms w.r.t CAN ID sequence length, it can also detect a wide variety of cyberattacks with an F1-score of between 0.81 and 0.99 (10.1109/AICCSA56895.2022.10017800)
  DOI : 10.1109/AICCSA56895.2022.10017800
• Learning Multi-Level Representations for Hierarchical Music Structure Analysis
  
  • Buisson Morgan
  • Mcfee Brian
  • Essid Slim
  • Crayencour Helene-Camille
  , 2022. Recent work in music structure analysis has shown the potential of deep features to highlight the underlying structure of music audio signals. Despite promising results achieved by such representations, dealing with the inherent hierarchical aspect of music structure remains a challenging problem. Because different levels of segmentation can be considered as equally valid, specifically designed representations should be optimized to improve hierarchical structure analysis. In this work, unsupervised learning of such representations using a contrastive approach operating at different timescales is explored. The proposed system is evaluated on flat and multi-level music segmentation. By leveraging both time and the hierarchical organization of music structure, we show that the obtained deep embeddings can encode meaningful patterns and improve segmentation at various levels of granularity.
• SSM-NET: FEATURE LEARNING FOR MUSIC STRUCTURE ANALYSIS USING A SELF-SIMILARITY-MATRIX BASED LOSS
  
  • Peeters Geoffroy
  • Angulo Florian
  , 2022. In this paper, we propose a new paradigm to learn audio features for Music Structure Analysis (MSA). We train a deep encoder to learn features such that the Self-Similarity-Matrix (SSM) resulting from those approximates a ground-truth SSM. This is done by minimizing a loss between both SSMs. Since this loss is differentiable w.r.t. its input features we can train the encoder in a straightforward way. We successfully demonstrate the use of this training paradigm using the Area Under the Curve ROC (AUC) on the RWC-Pop dataset.
• Exploiting device and audio data to tag music with User-Aware listening contexts
  
  • Ibrahim Karim M
  • V. Epure Elena
  • Peeters Geoffroy
  • Richard Gael
  , 2022. As music has become more available especially on music streaming platforms, people have started to have distinct preferences to fit to their varying listening situations, also known as context. Hence, there has been a growing interest in considering the user's situation when recommending music to users. Previous works have proposed user-aware autotaggers to infer situation-related tags from music content and user's global listening preferences. However, in a practical music retrieval system, the autotagger could be only used by assuming that the context class is explicitly provided by the user. In this work, for designing a fully automatised music retrieval system, we propose to disambiguate the user's listening information from their stream data. Namely, we propose a system which can generate a situational playlist for a user at a certain time 1) by leveraging user-aware music autotaggers, and 2) by automatically inferring the user's situation from stream data (e.g. device, network) and user's general profile information (e.g. age). Experiments show that such a context-aware personalized music retrieval system is feasible, but the performance decreases in the case of new users, new tracks or when the number of context classes increases.
• Rate Meta-distribution in mmW D2D Networks with Beam Misalignment
  
  • Quan Yibo
  • Coupechoux Marceau
  • Kelif Jean-Marc
  , 2022, pp.1825-1830. This paper studies the coverage performance of device-to-device (D2D) communication under the millimeter wave (mmW) spectrum. The transmitter and receiver sides of users are equipped with directional antennas and adopt beamforming (BF). By considering a truncated Gaussian misalignment assumption, we derive computationally tractable expressions of the conditional rate coverage probability's moments as a function of the number of antenna elements. The Beta approximation of the rate meta-distribution is obtained based on the first and the second moment. The numerical simulations confirm our analytical results. They show that the coverage performance can deteriorate significantly due to misalignment. Furthermore, an optimal number of antenna elements must be chosen to get the best coverage. In addition, there exists an optimal number of antennas which maximizes the number of users who satisfy the reliability constraints. This optimal value is a function of the reliability threshold. (10.1109/GLOBECOM48099.2022.10001302)
  DOI : 10.1109/GLOBECOM48099.2022.10001302
• TLS Early Data Resistance to Replay Attacks in Wireless Internet of Things
  
  • Kim Sung Yong
  • Goncharskyi Danylo
  • Gu Pengwenlong
  • Serhrouchni Ahmed
  • Khatoun Rida
  • Nait-Abdesselam Farid
  • Grund Jean-Jacques
  , 2022, pp.3539-3544. Transport Layer Security (TLS) is widely used for user authentication and encrypted data transmission in all kinds of networks. In its newly published version, TLS 1.3, a 0- RTT handshake protocol is proposed for session resumptions in low delay networks, which makes it possible to secure the data transmission and protect users from being monitored in wireless Internet of Things (IoTs). However, the 0-RTT TLS handshake protocol is vulnerable to the replay attack. In this paper, we propose a Time-Based One-Time Password (TOTP) empowered TLS encryption algorithm to resist replay attacks during the handshake process, in which we propose to integrate the TOTP into the encryption process of the EarlyData. It can significantly improve the forward secrecy of the 0-RTT handshake protocol and its capacity to resist the replay attack. On the other hand, we make no changes to the interaction process of the standardized 0- RTT handshake protocol to guarantee the compatibility of our proposed scheme, which makes our proposed scheme suitable for large area wireless IoTs. Simulation results show that under the premise of choosing an appropriate TOTP update rate, our proposed scheme can effectively resist replay attacks while ensuring the processing efficiency of the system. (10.1109/GLOBECOM48099.2022.10001106)
  DOI : 10.1109/GLOBECOM48099.2022.10001106
• Joint Coding of URLLC and eMBB in Wyner's Soft-Handoff Network in the Finite Blocklength Regime
  
  • Nikbakht Homa
  • Wigger Michele
  • Shitz Shlomo Shamai
  • Gorce Jean-Marie
  • Poor H Vincent
  , 2022, pp.1-6. Wyner's soft-handoff network is considered where transmitters simultaneously send messages of enhanced mobile broadband (eMBB) and ultra-reliable low-latency communication (URLLC) services. Due to the low-latency requirements, the URLLC messages are transmitted over fewer channel uses compared to the eMBB messages. To improve the reliability of the URLLC transmissions, we propose a coding scheme with finite blocklength codewords that exploits dirty-paper coding (DPC) to precancel the interference from eMBB transmissions. Rigorous bounds are derived for the error probabilities of eMBB and URLLC transmissions achieved by our scheme. Numerical results illustrate that they are lower than for standard time-sharing. (10.1109/globecom48099.2022.10000942)
  DOI : 10.1109/globecom48099.2022.10000942
• One Picture is Worth a Thousand Words: A New Wallet Recovery Process
  
  • Chabanne Herve
  • Despiegel Vincent
  • Guiga Linda
  , 2022, pp.1801-1806. We introduce a new wallet recovery pro-cess. Our solution associates 1) visual passwords: a photograph ofa secretly picked object (Chabanne et aI., 2013) with 2) ImageNet classifiers transforming images into binary vectors and, 3) obfuscated fuzzy matching (Galbraith and Zobernig, 2019) for the storage of visual passwords/retrieval of wallet seeds. Our experiments show that the replacement of long seed phrases by a photograph is possible (10.1109/GLOBECOM48099.2022.10001064)
  DOI : 10.1109/GLOBECOM48099.2022.10001064
• A survey on two-dimensional Error Correction Codes applied to fault-tolerant systems
  
  • Freitas David
  • Marcon César
  • Silveira Jarbas
  • Naviner Lirida
  • Mota João
  Microelectronics Reliability, Elsevier, 2022, 139, pp.114826. The number of memory faults operating in radiation environments increases with the electronic device miniaturization. One-dimensional (1D) Error Correction Codes (ECCs) are not efficient in mitigating these problems requiring two-dimensional (2D)-ECCs for providing superior error correction capacity with proportionally less energy and area consumption. The significant increase in publications in this area demands a study to guide and subsidize research decisions, mainly to determine a standardization method for comparing and evaluating ECCs. We propose a Systematic Literature Review (SLR) to investigate the most important features of 2D-ECCs used for mitigating faults in memories. This SLR revealed the most used ECCs, data size and redundancy overhead, encoder and decoder implementation technology, fault injection methods, and evaluation metrics. Besides, we extracted some ECC trends, such as reusing the encoder inside the decoder and targeting the three-dimensional (3D)-ECC to increase the error correction efficacy. The experimental results describe important research decisions of great value for this scientific community. (10.1016/j.microrel.2022.114826)
  DOI : 10.1016/j.microrel.2022.114826
• MIPNet: Neural Normal-to-Anisotropic-Roughness MIP mapping
  
  • Gauthier Alban
  • Faury Robin
  • Levallois Jérémy
  • Thonat Théo
  • Thiery Jean-Marc
  • Boubekeur Tamy
  ACM Transactions on Graphics, Association for Computing Machinery, 2022, 41 (6), pp.1-12. We present MIPNet, a novel approach for SVBRDF mipmapping which preserves material appearance under varying view distances and lighting conditions. As in classical mipmapping, our method explicitly encodes the multiscale appearance of materials in a SVBRDF mipmap pyramid. To do so, we use a tensor-based representation, coping with gradient-based optimization, for encoding anisotropy which is compatible with existing real-time rendering engines. Instead of relying on a simple texture patch average for each channel independently, we propose a cascaded architecture of multilayer perceptrons to approximate the material appearance using only the fixed material channels. Our neural model learns simple mipmapping filters using a differentiable rendering pipeline based on a rendering loss and is able to transfer signal from normal to anisotropic roughness. As a result, we obtain a drop-in replacement for standard material mipmapping, offering a significant improvement in appearance preservation while still boiling down to a single per-pixel mipmap texture fetch. We report extensive experiments on two distinct BRDF models. (10.1145/3550454.3555487)
  DOI : 10.1145/3550454.3555487
• Complexity reduction over Bi-RNN-based nonlinearity mitigation in dual-pol fiber-optic communications via a CRNN-based approach
  
  • Shahkarami Abtin
  • Yousefi Mansoor
  • Jaouën Yves
  Optical Fiber Technology, Elsevier, 2022, 74 (103072), pp.1-12. Bidirectional recurrent neural networks (bi-RNNs), in particular bidirectional long short term memory (bi-LSTM), bidirectional gated recurrent unit, and convolutional bi-LSTM models, have recently attracted attention for nonlinearity mitigation in fiber-optic communication. The recently adopted approaches based on these models, however, incur a high computational complexity which may impede their real-time functioning. In this paper, by addressing the sources of complexity in these methods, we propose a more efficient network architecture, where a convolutional neural network encoder and a unidirectional many-to-one vanilla RNN operate in tandem, each best capturing one set of channel impairments while compensating for the shortcomings of the other. We deploy this model in two different receiver configurations. In one, the neural network is placed after a linear equalization chain and is merely responsible for nonlinearity mitigation; in the other, the neural network is directly placed after the chromatic dispersion compensation and is responsible for joint nonlinearity and polarization mode dispersion compensation. For a 16-QAM 64 GBd dual-polarization optical transmission over 14 × 80 km standard single-mode fiber, we demonstrate that the proposed hybrid model achieves the bit error probability of the state-of-the-art bi-RNN-based methods with greater than 50% lower complexity, in both receiver configurations. (10.1016/j.yofte.2022.103072)
  DOI : 10.1016/j.yofte.2022.103072
• Local Decoding in Distributed Compression
  
  • Vatedka Shashank
  • Chandar Venkat
  • Tchamkerten Aslan
  IEEE Journal on Selected Areas in Information Theory, IEEE, 2022, pp.1-1. (10.1109/JSAIT.2023.3240187)
  DOI : 10.1109/JSAIT.2023.3240187
• Blockchain-Based Solution for Detecting and Preventing Fake Check Scams
  
  • Hammi Badis
  • Zeadally Sherali
  • Adja Yves Christian Elloh
  • Giudice Manlio Del
  • Nebhen Jamel
  IEEE Transactions on Engineering Management, Institute of Electrical and Electronics Engineers, 2022, 69 (6), pp.3710-3725. Fake check scam is one of the most common attacks used to commit fraud against consumers. This fraud is particularly costly for victims because they generally lose thousands of dollars as well as being exposed to judicial proceedings. Currently, there is no existing solution to authenticate checks and detect fake ones instantly. Instead, banks must wait for a period of more than 48 hours to detect the scam. In this context, we propose a blockchain-based scheme to authenticate checks and detect fake check scams. Moreover, our approach allows the revocation of used checks. More precisely, our approach helps the banks to share information about provided checks and used ones, without exposing the banks’ customers’ personal data. We demonstrate a proof of concept of our proposed approach using Namecoin and Hyperledger blockchain technologies. (10.1109/TEM.2021.3087112)
  DOI : 10.1109/TEM.2021.3087112
• Le barbier était une femme
  
  • Verneyre Séverine
  • Zayana Karim
  Tangente (Paris), 2022. S’il poursuit un but didactique en illustrant l’un des résultats les plus fondamentaux de la théorie des ensembles (le théorème de Cantor), le paradoxe du barbier risque cependant, sorti de son contexte, de se retourner contre vous. Aussi, après l’avoir remis en perspective – notamment de celle des programmes scolaires, tenterons-nous d’en dépasser le propos.
• Additional value of chest CT AI-based quantification of lung involvement in predicting death and ICU admission for COVID-19 patients
  
  • Galzin Eloise
  • Roche Laurent
  • Vlachomitrou Anna
  • Nempont Olivier
  • Carolus Heike
  • Schmidt-Richberg Alexander
  • Jin Peng
  • Rodrigues Pedro
  • Klinder Tobias
  • Richard Jean-Christophe
  • Tazarourte Karim
  • Douplat Marion
  • Sigal Alain
  • Bouscambert-Duchamp Maude
  • Si-Mohamed Salim Aymeric
  • Gouttard Sylvain
  • Mansuy Adeline
  • Talbot François
  • Pialat Jean-Baptiste
  • Rouvière Olivier
  • Milot Laurent
  • Cotton François
  • Douek Philippe
  • Duclos Antoine
  • Rabilloud Muriel
  • Boussel Loic
  Research in Diagnostic and Interventional Imaging, Elsevier, 2022, 4, pp.100018. (10.1016/j.redii.2022.100018)
  DOI : 10.1016/j.redii.2022.100018
• A secure cross-layer architecture for reactive routing in vehicle to vehicle (V2V) communications
  
  • Chbib Fadlallah
  • Zeadally Sherali
  • Khatoun Rida
  • Khoukhi Lyes
  • Fahs Walid
  • Haydar Jamal
  Vehicular Communications, Elsevier, 2022, 38, pp.100541. Vehicular communication is one of the essential technologies for increasing road safety, traffic efficiency, and comfort for pedestrians and drivers. In this context, the internet of vehicles is an emerging paradigm. However, with advances in vehicular communication, security threats have also emerged. Several vulnerabilities exist in vehicular communications, including Denial of Service (DoS), black hole attacks, and fabrication attacks. A malicious attack alters the packet information in a fabrication attack, causing congestion and high delays in the vehicular network. We propose two algorithms to protect the routing protocols in a vehicle-to-vehicle scenario against several attacks that target confidentiality, authentication, privacy, and integrity. The first algorithm detects the malicious behavior of each vehicle by calculating the percentage of modified destination addresses. If it exceeds a predetermined threshold, this vehicle is classified as malicious. Otherwise, it is a normal vehicle. The second algorithm detects malicious modifications based on the Signal to Interference Ratio (SIR) by monitoring the SIR value, adjusting the distance, altering the power received, and changing the transmitted power value. We performed simulations using the SUMO 0.22 simulator and Network Simulator (NS). The results obtained show an improvement in End-to-End (E2E) delay, Packet Delivery Ratio (PDR), and reduced overhead. (10.1016/j.vehcom.2022.100541)
  DOI : 10.1016/j.vehcom.2022.100541
• Exploiting modern GPUs architecture for real-time rendering of massive line sets
  
  • Schertzer Jérémie
  , 2022. In this thesis, we consider massive line sets generated from brain tractograms. They describe neural connections that are represented with millions of poly-line fibers, summing up to billions of segments. Thanks to the two-staged mesh shader pipeline, we build a tractogram renderer surpassing state-of-the-art performances by two orders of magnitude.Our performances come from fiblets: a compressed representation of segment blocks. By combining temporal coherence and morphological dilation on the z-buffer, we define a fast occlusion culling test for fiblets. Thanks to our heavily-optimized parallel decompression algorithm, surviving fiblets are swiftly synthesized to poly-lines. We also showcase how our fiblet pipeline speeds-up advanced tractogram interaction features.For the general case of line rendering, we propose morphological marching: a screen-space technique rendering custom-width tubes from the thin rasterized lines of the G-buffer. By approximating a tube as the union of spheres densely distributed along its axes, each sphere shading each pixel is retrieved relying on a multi-pass neighborhood propagation filter. Accelerated by the compute pipeline, we reach real-time performances for the rendering of depth-dependant wide lines.To conclude our work, we implement a virtual reality prototype combining fiblets and morphological marching. It makes possible for the first time the immersive visualization of huge tractograms with fast shading of thick fibers, thus paving the way for diverse perspectives.
• Learning anatomical digital twins in pediatric 3D imaging for renal cancer surgery
  
  • La Barbera Giammarco
  , 2022. Pediatric renal cancers account for 9% of pediatric cancers with a 9/10 survival rate at the expense of the loss of a kidney. Nephron-sparing surgery (NSS, partial removal of the kidney) is possible if the cancer meets specific criteria (regarding volume, location and extent of the lesion). Indication for NSS is relying on preoperative imaging, in particular X-ray Computerized Tomography (CT). While assessing all criteria in 2D images is not always easy nor even feasible, 3D patient-specific models offer a promising solution. Building 3D models of the renal tumor anatomy based on segmentation is widely developed in adults but not in children. There is a need of dedicated image processing methods for pediatric patients due to the specificities of the images with respect to adults and to heterogeneity in pose and size of the structures (subjects going from few days of age to 16 years). Moreover, in CT images, injection of contrast agent (contrast-enhanced CT, ceCT) is often used to facilitate the identification of the interface between different tissues and structures but this might lead to heterogeneity in contrast and brightness of some anatomical structures, even among patients of the same medical database (i.e., same acquisition procedure). This can complicate the following analyses, such as segmentation. The first objective of this thesis is to perform organ/tumor segmentation from abdominal-visceral ceCT images. An individual 3D patient model is then derived. Transfer learning approaches (from adult data to children images) are proposed to improve state-of-the-art performances. The first question we want to answer is if such methods are feasible, despite the obvious structural difference between the datasets, thanks to geometric domain adaptation. A second question is if the standard techniques of data augmentation can be replaced by data homogenization techniques using Spatial Transformer Networks (STN), improving training time, memory requirement and performances. In order to deal with variability in contrast medium diffusion, a second objective is to perform a cross-domain CT image translation from ceCT to contrast-free CT (CT) and vice-versa, using Cycle Generative Adversarial Network (CycleGAN). In fact, the combined use of ceCT and CT images can improve the segmentation performances on certain anatomical structures in ceCT, but at the cost of a double radiation exposure. To limit the radiation dose, generative models could be used to synthesize one modality, instead of acquiring it. We present an extension of CycleGAN to generate such images, from unpaired databases. Anatomical constraints are introduced by automatically selecting the region of interest and by using the score of a Self-Supervised Body Regressor, improving the selection of anatomically-paired images between the two domains (CT and ceCT) and enforcing anatomical consistency. A third objective of this work is to complete the 3D model of patient affected by renal tumor including also arteries, veins and collecting system (i.e. ureters). An extensive study and benchmarking of the literature on anatomic tubular structure segmentation is presented. Modifications to state-of-the-art methods for our specific application are also proposed. Moreover, we present for the first time the use of the so-called vesselness function as loss function for training a segmentation network. We demonstrate that combining eigenvalue information with structural and voxel-wise information of other loss functions results in an improvement in performance. Eventually, a tool developed for using the proposed methods in a real clinical setting is shown as well as a clinical study to further evaluate the benefits of using 3D models in pre-operative planning. The intent of this research is to demonstrate through a retrospective evaluation of experts how criteria for NSS are more likely to be found in 3D compared to 2D images. This study is still ongoing.
• Towards a Formal Verification of Attack Graphs
  
  • Catta Davide
  • Leneutre Jean
  • Malvone Vadim
  , 2022, 3345. In this perspective paper, we propose different formalizations of games that are played over Attack Graphs between an Attacker and a Defender. In all such games we propose a formal approach (such as logics and automata theory) to check whether the Attacker has a strategy to win the game.
• Repeated Augmented Rehearsal : A simple but strong baseline for online continual learning
  
  • Zhang Yaqian
  • Pfahringer Bernhard
  • Frank Eibe
  • Bifet Albert
  • Lim Nick Jin Sean
  • Jia Yunzhe
  , 2022. Online continual learning (OCL) aims to train neural networks incrementally from a non-stationary data stream with a single pass through data. Rehearsal-based methods attempt to approximate the observed input distributions over time with a small memory and revisit them later to avoid forgetting. Despite their strong empirical performance, rehearsal methods still suffer from a poor approximation of past data’s loss landscape with memory samples. This paper revisits the rehearsal dynamics in online settings. We provide theoretical insights on the inherent memory overfitting risk from the viewpoint of biased and dynamic empirical risk minimization, and examine the merits and limits of repeated rehearsal.Inspired by our analysis, a simple and intuitive baseline, repeated augmented rehearsal (RAR), is designed to address the underfitting-overfitting dilemma of online rehearsal. Surprisingly, across four rather different OCL benchmarks,this simple baseline outperforms vanilla rehearsal by 9\%-17\% and also significantly improves the state-of-the-art rehearsal-based methods MIR, ASER, and SCR. We also demonstrate that RAR successfully achieves an accurate approximation of the loss landscape of past data and high-loss ridge aversion in its learning trajectory. Extensive ablation studies are conducted to study the interplay between repeated and augmented rehearsal, and reinforcement learning (RL) is applied to dynamically adjust the hyperparameters of RAR to balance the stability-plasticity trade-off online.
• Design and implementation of high efficiency power amplifiers for 5G Applications
  
  • Bachi Joe
  , 2022. The increasing complexity of modulation schemes brought on by the evolution of mobile communication standards has led to high peak to average power ratio (PAPR) signals. As a result, traditional linear power amplifier (PA) architectures are no longer suitable as they exhibit low average efficiency when operating with such signals. One of the possible solutions to this issue is load modulation-based architectures which are capable of providing higher average efficiency. This work focuses on the analysis, design, and implementation of the two main load modulation architectures: Outphasing (OPA) and Doherty (DPA). The Outphasing architecture is studied under its different forms and a new unified design method is proposed for OPA combiners. A second analysis is conducted on DPA combiners, resulting in a new analysis method capable of determining the maximum back-off achievable by a given combiner architecture in Doherty mode. Unlike existing works, the proposed method also determines the required driving currents at the inputs of the combiner to maintain ideal Doherty conditions throughout the Doherty region. In order to validate this technique, a twostage class-E DPA with compact LC combiner is designed and implemented using 130nm RF-SOI. Measured performance is in-line with the state of the art as the PA achieves a peak PAE of 51% at 32dBm output power under 3.4V supply voltage at 2.3GHz in CW mode. From 2.1GHz to 2.5GHz, the PA shows an average output power and PAE higher than 26.9dBm and 39% respectively at -35dBc E-UTRA ACLR when using a 10MHz50RB QPSK LTE uplink signal with memoryless digital predistortion (DPD). At 2.3GHz, the PA achieves a linear Pout and PAE of 28.85dBm and 42.8% respectively. Next, a system analysis is performed on the Outphasing transmitter system (OTX) which contains both the RF OPA as well as the signal processing interface and analog interface known as the signal component separator (SCS). The design and operation of OPA in both class-B and class-E is studied resulting in a dual-input class-E OPA design. Different DPD architectures are studied including the look-up table DPD and the behavioural modelling-based architectures. Finally, an IN-SCS DPD architecture is put forward as a potential novel solution allowing the integration of the DPD block into the SCS providing abasis for future research.