2026Other
Dynamic Layer Selection for Efficient Tone Recognition in Self-Supervised Speech Models
Saint Germes B. Bengono Obiang, Norbert Tsopzé, Paulin Melatagia Yonta
Checking session…
INF
Chargé de Cours
Scientific activity
Researcher profile
Faculty researcher at the University of Yaoundé I. Grade: Chargé de Cours.
Scientific publications
Saint Germes B. Bengono Obiang, Norbert Tsopzé, Paulin Melatagia Yonta
Ingrid Pamela Nguemkam Tebou, Norbert Tsopzé, Dieudonné Tchuente
Information Systems Frontiers
Claude Kanyou, Etienne Kouokam, Norbert Tsopzé
Theory in Biosciences
Mbietieu Amos Mb., T. M., Kouamou Georges Edouard, Norbert Tsopzé
Science of Remote Sensing
Advancements in remote sensing and artificial intelligence are transforming maritime object detection in High-Resolution Satellite Imagery (HRSI), with critical applications in environmental monitoring, maritime security, and sustainable ocean management. However, detecting ships in HRSI remains challenging due to variations in object size, orientation, and environmental conditions. This study introduces Ship Feature Pyramid Network (ShipFPN), a novel deep learning architecture that integrates spectral convolutions, adaptive feature pyramids, and robust data augmentation techniques to enhance detection accuracy and generalizability. ShipFPN is evaluated on the ShipRSImageNet dataset, demonstrating significant improvements over existing FPN-based models in precision, recall, and Generalized Intersection over Union (GIoU). While this study focuses on ship detection, ShipFPN’s architecture is designed to be adaptable for detecting other maritime objects. By improving the accuracy and robustness of ship detection, this work contributes to enhanced marine traffic monitoring, environmental protection, and climate resilience in satellite-based remote sensing. • ShipFPN improves detection of small ships in high-resolution satellite imagery. • Our FPN architecture uses adaptive fusion and attention-based enhancement. • The model achieves higher mAP on ShipRSImageNet than classic FPN baselines.
Loic Youmbi, Ali Wacka, Norbert Tsopzé
Communications in computer and information science
Ingrid Pamela Nguemkam Tebou, Norbert Tsopzé, Dieudonné Tchuente
Communications in computer and information science
Michaël Chirmeni Boujike, Jerry Lonlac, Norbert Tsopzé, Engelbert Mephu Nguifo, Laure Pauline Fotso
International Journal of General Systems
The traditional algorithms that extract the gradual patterns often face the problem of managing the quantity of mined patterns, and in many applications, the calculation of all these patterns can prove to be intractable for the user-defined frequency threshold. Moreover, the concept of gradualness is defined just as an increase or a decrease variation. Indeed, a gradualness is considered as soon as the values of the attribute on both objects are different. This does not take into account the level of variation. Then, the variation of 10−6 is considered as the same way as that of 106. As a result, numerous quantities of patterns extracted by traditional algorithms can be presented to the user, although their gradualness (due to the small variation) could be only a noise in the data. To address this issue, this paper suggests introducing the gradualness threshold from which to consider an increase or a decrease variation. In contrast to the literature approaches, the proposed approach takes into account the user's preferences on the gradualness threshold. The user knowledge could be used to fix the value of gradualness threshold. The proposed algorithm makes it possible to extract gradual patterns on certain databases where state-of-the-art gradual patterns mining algorithms fail due to too large search space. Moreover, results from an experimental evaluation on real databases show that the proposed algorithm is scalable, efficient, and can eliminate numerous patterns that do not verify specific gradualness requirements to show a small set of patterns to the user.
Cezar Mbiethieu, Norbert Tsopzé, Engelbert Mephu Nguifo
HAL (Le Centre pour la Communication Scientifique Directe)
submission to Episciences
Florentin Jiechieu, Norbert Tsopzé
HAL (Le Centre pour la Communication Scientifique Directe)
International audience
Norbert Tsopzé, Félicité Gamgne Domgue
Information Sciences
Félicité Gamgne Domgue, Norbert Tsopzé, René Ndoundam
Social Network Analysis and Mining
Félicité Gamgne Domgue, Norbert Tsopzé, René Ndoundam
International Journal of General Systems
Community detection in directed networks appears as one of the most relevant topics in the field of network analysis. One of the common themes in its formalizations is information flow clustering in a network. Such clusters can be extracted by using triads, expected to play an important role in the detection of that type of communities since communities could be centered round core nodes called kernels. Triads in directed graphs are directed sub-graphs of three nodes involving at least two links between them. To identify communities in directed networks, this paper proposes an in-seed-centric scheme based on directed triads. We also propose a new metric of the communities' quality based on the triad density of communities. To validate our approach, an experiment was conducted on some networks showing it has better performance on triad-based density over some state-of-the-art methods.
Florentin Flambeau Jiechieu Kameni, Norbert Tsopzé
Revue Africaine de la Recherche en Informatique et Mathématiques Appliquées
The aim of this work is to use a hybrid approach to extract CVs' competences. The extraction approach for competences is made of two phases: a segmentation into sections phase within which the terms representing the competences are extracted from a CV; and a prediction phase that consists from the features previously extracted, to foretell a set of competences that would have been deduced and that would not have been necessary to mention in the resume of that expert. The main contributions of the work are two folds : the use of the approach of the hierarchical clustering of a résume in section before extracting the competences; the use of the multi-label learning model based on SVMs so as to foretell among a set of skills, those that we deduce during the reading of a CV. Experimentation carried out on a set of CVs collected from an internet source have shown that, more than 10% improvement in the identification of blocs compared to a model of the start of the art. The multi-label competences model of prediction allows finding the list of competences with a precision and a reminder respectively in an order of 90.5 % and 92.3 %. . L’objectif de ce travail est d’utiliser l’approche héirarchique d’extraction des informations dans le CV pour en extraire les compétences. L’approche d’extraction des compétences proposée s’effectue en deux grandes phases : une phase de segmentation du CV en sections classées suivant leurs contenus et à partir desquelles les termes représentant les compétences (compétences de bases) sont extraits; et une phase de prédiction qui consiste à partir des caractéristiques extraites précédemment, à prédire un ensemble de compétences qu’un expert aurait déduites, et que ces compétences ne seraient pas nécessairement mentionnées dans le CV (compétences implicites). Les principales contributions de ce travail sont : l’utilisation de l’approche hiérarchique de segmentation du CV en sections pour extraire les compétences dans le CV; l’amélioration de la l’approche de segmentation des CV; enfin, l’utilisation de l’approche binary relevance de classification multi-label pour prédire les compétences implicites du CV. Les expérimentations effectuées sur un jeu de CVs collectés sur Internet ont montré une amélioration de la précision de l’identification des blocs de plus de 10% comparé à un modèle de l’état de l’art. Aussi, le modèle de prédiction multi-label des compétences, permet de retrouver la liste des compétences avec une précision et un rappel respectivement de l’ordre de 90,5% et 92,3%.
Lauraine Tiogning Kueti, Norbert Tsopzé, Cezar Mbiethieu, Engelbert Mephu Nguifo, Laure Pauline Fotso
International Journal of General Systems
We propose a novel approach to define Artificial Neural Network(ANN) architecture from Boolean factors. ANNs are a subfield of machine learning applicable to several areas of life. However, defining its architecture for solving a given problem is not formalized and remains an open research problem. Since it is difficult to look into the network and figure out exactly what it has learnt, the complexity of such a technique makes its interpretation more tedious. We propose in this paper to build feedforward ANNs using the optimal factors obtained from the Boolean context representing a data. Since optimal factors completely cover the data and therefore give an explanation to these data, We could give an interpretation to the neurons activation and justify the presence of a neuron in our proposed neural network. We show through experiments and comparisons on the use data sets that this approach provides relatively better results for some key performance measures.
Félicité Gamgne Domgue, Norbert Tsopzé, Arnaud Ahouandjinou
Association en Recherche d’Information et Applications
RÉSUMÉ. La détection des communautés est devenue un domaine de recherche majeur ces dernières années. Plusieurs algorithmes appliqués aux graphes orientés ont été developpés. Ces derniers se focalisent sur la densité de liens à l'intérieur des communautés et considèrent la relation entre les noeuds comme symmétrique, car ils ignorent l'orientation des liens, ce qui biaise les résultats en produisant des communautés non-significatives. Ce document propose un algorithme basé sur l'extraction des kernels via la distribution des triades, utilisant l'optimi- sation de la nouvelle métrique Kernel Degree Clustering (KDC), et trouve des communautés plus sémantiques que la modularité, en accord à la notion de centralisation de l'information. Les expérimentations montrent que la nouvelle approche produit les résultats préconisés que ceux produits par certains algorithmes de détection de communautés de l'état de l'art.
Lauraine Tiogning Kueti, Norbert Tsopzé, Cezar Mbiethieu, Engelbert Mephu Nguifo, Laure Pauline Fotso
Due to its ability to solve nonlinear problems, Artificial Neural Network (ANN) could be applied in several areas of life. However, defining its architecture for solving a given problem is not formalized and remains an open research problem. On the other hand the complexity of such a technique due to its “black box” aspect, makes its interpretation more tedious. Since optimal factors completely cover the data and therefore give an explanation to these data, we propose in this paper to build feedforward ANNs using the optimal factors obtained from the boolean context representing a data. We show through experiments and comparisons on the use datasets that this approach provides relatively better results than those existing in the literature.
Norbert Tsopzé, Gamgne Domgue Félicité
HAL (Le Centre pour la Communication Scientifique Directe)
Community detection in directed networks appears as one of dominant research works in network analysis. Most existing models for community detection are symmetric, in which incom-ming and outgoing links are treated equally. In this paper, we propose a method based on Formal Concepts that takes into account both symmetric and non-symmetric properties of links. To validate our approach, experiments on known networks with symmetric property of links in directed graphs show that outputs detect the same communities as some of state-of-art methods; while, with non-symmetric property of links, only this approach uncovers expected communities. Its Experimental results on benchmark are also better than those of many existing algorithms.
Mickaël Coustaty, Norbert Tsopzé, Alain Bouju, Karell Bertet, Georges Louis
Applied Ontology
A huge number of historical documents have been digitized over the last ten years. Browsing into these collections can be done using query by keywords or query by example systems. Going from one kind of query to another raises the problem of the semantic gap. In order to deal with this problem, thi s paper presents an ontology-based approach to the resolution of the semantic gap problem that uses inference rules with historical images. To do this, historians’ knowledge and knowledge from the document processing domain were modeled using dedicated ontologies. Then, links between the regions of interest from the computer vision algorithms on the one hand, and their meaning on the other hand, were automatically created. These links will subsequently be used to help historians retrieve similar images. Based on the three ontologies defined and combined in this approach, we have defined rules to automatically annotate an image (to define the background for example) or a part of an image (to identify a letter, a body-part, …).
Cyrine Arouri, Engelbert Mephu Nguifo, Sabeur Aridhi, C. Roucelle, Gaëlle Bonnet-Loosli, Norbert Tsopzé
arXiv (Cornell University)
The choice of architecture of artificial neuron network (ANN) is still a challenging task that users face every time. It greatly affects the accuracy of the built network. In fact there is no optimal method that is applicable to various implementations at the same time. In this paper we propose a method to construct ANN based on clustering, that resolves the problems of random and ad hoc approaches for multilayer ANN architecture. Our method can be applied to regression problems. Experimental results obtained with different datasets, reveals the efficiency of our method.
Cyrine Arouri, Engelbert Mephu Nguifo, Sabeur Aridhi, C. Roucelle, Bonnet-Loosli, Gaëlle, Norbert Tsopzé
HAL (Le Centre pour la Communication Scientifique Directe)
International audience
Christophe Rigaud, Norbert Tsopzé, Jean-Christophe Burie, Jean-Marc Ogier
HAL (Le Centre pour la Communication Scientifique Directe)
Christophe Rigaud, Norbert Tsopzé, Jean-Christophe Burie, Jean-Marc Ogier
Lecture notes in computer science
Christophe Rigaud, Norbert Tsopzé, Jean‐Christophe Burie, Jean-Marc Ogier
HAL (Le Centre pour la Communication Scientifique Directe)
National audience
Mickaël Coustaty, Norbert Tsopzé, Alain Bouju, Karell Bertet, Georges Louis, Jean-Marc Ogier
HAL (Le Centre pour la Communication Scientifique Directe)
International audience
Norbert Tsopzé, Engelbert Mephu Nguifo, Gilbert Tindo
Artificial Neural Networks classifiers have many advantages such as: noise tolerance, possibility of parallelization, better training with a small quantity of data.... Coupling neural networks with an explanation component will increase its usage
Norbert Tsopzé, Engelbert Mephu Nguifo, Gilbert Tindo
The current development of knowledge discovery domain has pointed out a high number of applications where the need of explanation is at the heart of the process. Using neural networks for those applications requires to be able to provide a set of rules extracted from the trained neural networks, that can help the user to comprehend the learning process. The current literature reports two kinds of rules: `if condition then conclusion' (called if-then) and `if m of conditions then conclusion' (also called MofN). We propose a new method able to extract one intermediate structure (called generators list) from which it is possible to extract both forms of rules. The extracted structure is a generic representation that gives the possibility to the user to visualize each form of rules extracted from the multilayer artificial neural networks.
Les réseaux de neurones artificiels connaissent des succès dans plusieurs domaines. Maisles utilisateurs des réseaux de neurones sont souvent confrontés aux problèmes de définitionde son architecture et d’interprétabilité de ses résultats. Plusieurs travaux ont essayé d’apporterune solution à ces problèmes. Pour les problèmes d’architecture, certains auteurs proposentde déduire cette architecture à partir d’un ensemble de connaissances décrivant le domaine duproblème et d’autres proposent d’ajouter de manière incrémentale les neurones à un réseauayant une taille initiale minimale. Les solutions proposées pour le problème d’interprétabilitédes résultats consistent à extraire un ensemble de règles décrivant le fonctionnement du réseau.Cette thèse contribue à la résolution de ces deux problèmes. Nous nous limitons à l’utilisationdes réseaux de neurones dans la résolution des problèmes de classification.Nous présentons dans cette thèse un état de l’art des méthodes existantes de recherche d’architecturede réseaux de neurones : une étude théorique et expérimentale est aussi faite. Decette étude, nous observons comme limites de ces méthodes la disponibilité absolue des connaissancespour construire un réseau interprétable et la construction des réseaux difficiles à interpréteren absence de connaissances. En alternative, nous proposons une méthode appelée CLANN(Concept Lattice-based Artificial Neural network) basée les treillis de Galois qui construit undemi-treillis à partir des données et déduire de ce demi-treillis l’architacture du réseau. CLANNétant limitée à la résolution des problèmes à deux classes, nous proposons MCLANN permettantd’étendre cette méthodes de recherche d’architecture des réseaux de neurones aux problèmes àplusieurs classes.Nous proposons aussi une méthode appelée ’Approche des MaxSubsets’ pour l’extractiondes règles à partir d’un réseau de neurones. La particularité de cette méthode est la possibilitéd’extraire les deux formats de règles (’si alors’ et ’m parmi N’) à partir d’une structure quenous construisons. Nous proposons aussi une façon d’expliquer le résultat calculé par le réseauconstruit par la méthode MCLANN au sujet d’un exemple.
Engelbert Mephu Nguifo, Norbert Tsopzé, Gilbert Tindo
Studies in computational intelligence
Engelbert Mephu Nguifo, Norbert Tsopzé, Gilbert Tindo
Lecture notes in computer science
Norbert Tsopzé, Engelbert Mephu Nguifo, G. Tindo
HAL (Le Centre pour la Communication Scientifique Directe)
12 pages
Norbert Tsopzé, Engelbert Mephu Nguifo, Gilbert Tindo
Abstract. Multi-layer neural networks have been successfully applied in a wide range of supervised and unsupervised learning applications. As they often produce incomprehensible models they are not widely used in data mining applications. To avoid such limitations, comprehensive models have been previously introduced making use of an apriori knowledge to build the network architecture. They permit to neural network methods to deserve a place in the tool boxes of data mining specialists. However, as the apriori knowledge is not always available for every new dataset, we hereby propose a novel approach that generates a concept semi-lattice from initial dataset, to directly build the neural network architecture. Carried out experiments showed the soundness and efficiency of our approach on various UCI. 1
Cezar Mbiethieu, Norbert Tsopzé, Engelbert Mephu Nguifo
Computer Vision and Image Understanding
Saint Germes Bienvenu Bengono Obiang, Norbert Tsopzé, Paulin Melatagia Yonta, Jean-François Bonastre, T. Jimenez
ACM Transactions on Asian and Low-Resource Language Information Processing
Many sub-Saharan African languages are categorized as tone languages, and for the most part, they are classified as low-resource languages due to the limited resources and tools available to process these languages. Identifying the tone associated with a syllable is therefore a key challenge for speech recognition in these languages. We propose models that automate the recognition of tones in continuous speech that can easily be incorporated into a speech recognition pipeline for these languages. We have investigated different neural architectures as well as several feature extraction algorithms in speech (FBs (Filter Banks), LEAF (Learnable Frontend), CS (Cestrogram), MFCC (Mel-Frequency Cepstral Coefficients)). In the context of low-resource languages, we also evaluated W2V (Wav2vec 2.0) models for this task. In this work, we use a public speech recognition dataset on Yoruba. As for the results, using the combination of features obtained from CS and FBs, we obtain a minimum TER (Tone Error Rate) of 19.54%, whereas for the evaluations of the models using W2V, we have a TER of 17.72%, demonstrating that the use of W2V provides better performance than the models used in the literature for tone identification on low-resource languages.
Félicité Gamgne Domgue, Norbert Tsopzé, René Ndoundam
Knowledge and Information Systems
Félicité Gamgne Domgue, Norbert Tsopzé, René Ndoundam
Research Square
Abstract Many hierarchical methods for community detection in multicolored networks are capable of finding clusters when there are interslice correlation between layers. However, in general, they aggregate all the links in different layer treating them as being equivalent. Therefore, such aggregation might ignore the information about the relevance of a dimension in which the node is involved. In this paper we fill this gap by proposing a hierarchical classification based-Louvain method for interslice-multicolored networks. In particular we define a new node centrality measure named \textit{Attractivity} to describe the inter-slice correlation that incorporates within and across-dimension topological features in order to identify the relevant dimension. Then, after merging dimensions through a frequential aggregation, we group nodes by their relational and attribute similarity, where attributes correspond to their relevant dimensions. We conduct an extensive experimentation using seven real-world multicolored networks, which also includes comparison with state-of-the-art methods. Results show the significance of our proposed method in discovering relevant communities over multiple dimensions and highlight its ability in producing optimal covers with higher values of the multidimensional version of the modularity function.
Maureen Domche, Jerry Lonlac, Norbert Tsopzé, Engelbert Mephu Nguifo
HAL (Le Centre pour la Communication Scientifique Directe)
Gradual patterns mining aims to extract from numerical data, the frequent covariations between attributes of the form "The more/less x1, ..., the more/less xn". It is significant for capturing the variability of numerical values in applications when the volume of data becomes large. Moreover, statistical correlation, which highlights relationships between variables, can also be used to express covariations in numerical data.Although gradual patterns and statistical correlations capture covariations from numerical data, gradual patterns provide more expressive knowledge. To our knowledge, no work in the literature focused on studying the differences between these two concepts to highlight the limits and the advantages of gradual patterns regarding statistical correlations. In this article, we conduct a comparative study between the gradual patterns extracted using different semantics of graduality and statistical correlations, presenting the similarities, differences, advantages and disadvantages of each of the concepts for numerical data processing. This study is completed by experiments carried out on several numerical databases, the results of which confirm the contribution of gradual patterns compared to statistical correlations.
Saint Germes BENGONO OBIANG, Norbert Tsopzé, Paulin Melatagia Yonta, Jean-François Bonastre, Tania Jiménez
SSRN Electronic Journal
Audrey Fongue, Jerry Lonlac, Norbert Tsopzé
Lecture notes in computer science
Michaël Chirmeni Boujike, Norbert Tsopzé, Jerry Lonlac, Engelbert Mephu Nguifo, Rosette Nganmeni Njamnou, Laure Pauline Fotso
HAL (Le Centre pour la Communication Scientifique Directe)
International audience
Saint Germes Bienvenu Bengono Obiang, Norbert Tsopzé
L'analyse des opinions consiste a extraire des connaissances a partir des commentaires laisses par les utilisateurs a propos d'un produit, service, texte,... L'analyse des opinions basee sur les aspects consiste alors a decomposer le commentaire afin d'extraire les aspects que cet utilisateur a evalue. Le modele propose par Jebbara et Cimiano, vainqueur de la competition SemEval2016 n'extrait pas correctement des aspects composes et ne prend pas en compte les caracteristiques lexico-grammaticales des textes en entree, ce qui limite aussi ses performances dans la detection des aspects. Nous proposons une amelioration du modele de Jebbara et Cimiano. en y introduisant des unites CRF afin de prendre en compte les dependances entre les etiquettes et ajoutant aux entrees du modele des caracteristiques lexico-grammaticales. Les experimentations faites sur les deux jeux de donnees de SemEval2016 ont permis de tester cette approche et montrer une amelioration de la mesure F-score d'environ 3.5%. ABSTRACT. The Internet contains a wealth of information in the form of unstructured texts such as customer comments on products, events and more. By extracting and analyzing the opinions expressed in customer comments in detail, it is possible to obtain valuable opportunities and information for customers and companies. The model proposed by Jebbara and Cimiano. for the extraction of aspects, winner of the SemEval2016 competition, suffers from the absence of lexico-grammatic input characteristics and poor performance in the detection of compound aspects. We propose the model based on a recurrent neural network for the task of extracting aspects of an entity for sentiment analysis. The proposed model is an improvement of the Jebbara and Cimiano model. The modification consists in adding a CRF to take into account the dependencies between labels and we have extended the characteristics space by adding grammatical level characteristics and lexical level characteristics. Experiments on the two SemEval2016 data sets tested our approach and showed an improvement in the F-score measurement of about 3.5%.
Saint Germes Bienvenu Bengono Obiang, Norbert Tsopzé
Revue Africaine de la Recherche en Informatique et Mathématiques Appliquées
The Internet contains a wealth of information in the form of unstructured texts such as customer comments on products, events and more. By extracting and analyzing the opinions expressed in customer comments in detail, it is possible to obtain valuable opportunities and information for customers and companies. The model proposed by Jebbara and Cimiano. for the extraction of aspects, winner of the SemEval2016 competition, suffers from the absence of lexico-grammatic input characteristics and poor performance in the detection of compound aspects. We propose the model based on a recurrent neural network for the task of extracting aspects of an entity for sentiment analysis. The proposed model is an improvement of the Jebbara and Cimiano model. The modification consists in adding a CRF to take into account the dependencies between labels and we have extended the characteristics space by adding grammatical level characteristics and lexical level characteristics. Experiments on the two SemEval2016 data sets tested our approach and showed an improvement in the F-score measurement of about 3.5%. L'analyse des opinions consiste à extraire des connaissances à partir des commentaires laissés par les utilisateurs à propos d'un produit, service, texte,... L'analyse des opinions basée sur les aspects consiste alors à décomposer le commentaire afin d'extraire les aspects que cet utilisateur a évalué. Le modèle proposé par Jebbara et Cimiano, vainqueur de la compétition SemEval2016 n'extrait pas correctement des aspects composés et ne prend pas en compte les caractéristiques lexico-grammaticales des textes en entrée, ce qui limite aussi ses performances dans la détection des aspects. Nous proposons une amélioration du modèle de Jebbara et Cimiano. en y introduisant des unités CRF afin de prendre en compte les dépendances entre les étiquettes et ajoutant aux entrées du modèle des caractéristiques lexico-grammaticales. Les expérimentations faites sur les deux jeux de données de SemEval2016 ont permis de tester cette approche et montrer une amélioration de la mesure F-score d'environ 3.5%. ABSTRACT. The Internet contains a wealth of information in the form of unstructured texts such as customer comments on products, events and more. By extracting and analyzing the opinions expressed in customer comments in detail, it is possible to obtain valuable opportunities and information for customers and companies. The model proposed by Jebbara and Cimiano. for the extraction of aspects, winner of the SemEval2016 competition, suffers from the absence of lexico-grammatic input characteristics and poor performance in the detection of compound aspects. We propose the model based on a recurrent neural network for the task of extracting aspects of an entity for sentiment analysis. The proposed model is an improvement of the Jebbara and Cimiano model. The modification consists in adding a CRF to take into account the dependencies between labels and we have extended the characteristics space by adding grammatical level characteristics and lexical level characteristics. Experiments on the two SemEval2016 data sets tested our approach and showed an improvement in the F-score measurement of about 3.5%.
Michaël Chirmeni Boujike, Jerry Lonlac, Norbert Tsopzé, Engelbert Mephu Nguifo
HAL (Le Centre pour la Communication Scientifique Directe)
International audience
Stanley Onyekachi Uche, Norbert Tsopzé, Deborah Ebem
Advances in Multidisciplinary & Scientific Research Journal Publication
Terrorism has long been a major threat to the world for many years and different governments have used different approaches to tackle it. This study reports on the use of available data about terrorist incidents all around the world in combating terrorism with the application of deep learning. There was a comprehensive literature review and the analysis of existing systems and ideas gathered were used to develop the system. This project is done in order to improve on the work carried out by Trisha J. (2018). To improve on the work, extra features were introduced in the dataset and a deep neural network (DNN) model for predicting the success of terrorist attacks was developed. Dataset from the Global Terrorism Database (GTD) were used to train the model. Our proposed model achieved performance accuracy of 91.371% as opposed to that of Trisha J. (2018) which achieved the performance accuracy of 91.18%. Keywords: Terrorism, deep neural network, global terrorism database, data mining
Michaël Chirmeni Boujike, Jerry Lonlac, Norbert Tsopzé, Engelbert Mephu Nguifo
arXiv (Cornell University)
The gradual patterns that model the complex co-variations of attributes of the form "The more/less X, The more/less Y" play a crucial role in many real world applications where the amount of numerical data to manage is important, this is the biological data. Recently, these types of patterns have caught the attention of the data mining community, where several methods have been defined to automatically extract and manage these patterns from different data models. However, these methods are often faced the problem of managing the quantity of mined patterns, and in many practical applications, the calculation of all these patterns can prove to be intractable for the user-defined frequency threshold and the lack of focus leads to generating huge collections of patterns. Moreover another problem with the traditional approaches is that the concept of gradualness is defined just as an increase or a decrease. Indeed, a gradualness is considered as soon as the values of the attribute on both objects are different. As a result, numerous quantities of patterns extracted by traditional algorithms can be presented to the user although their gradualness is only a noise effect in the data. To address this issue, this paper suggests to introduce the gradualness thresholds from which to consider an increase or a decrease. In contrast to literature approaches, the proposed approach takes into account the distribution of attribute values, as well as the user's preferences on the gradualness threshold and makes it possible to extract gradual patterns on certain databases where literature approaches fail due to too large search space. Moreover, results from an experimental evaluation on real databases show that the proposed algorithm is scalable, efficient, and can eliminate numerous patterns that do not verify specific gradualness requirements to show a small set of patterns to the user.
Saint Germes Bienvenu Bengono Obiang, Norbert Tsopzé
Félicité Gamgne Domgue, Norbert Tsopzé, René Ndoundam
HAL (Le Centre pour la Communication Scientifique Directe)
Jiechieu Kameni Florentin Flambeau, Norbert Tsopzé
arXiv (Cornell University)
During the last decade, deep neural networks (DNN) have demonstrated impressive performances solving a wide range of problems in various domains such as medicine, finance, law, etc. Despite their great performances, they have long been considered as black-box systems, providing good results without being able to explain them. However, the inability to explain a system decision presents a serious risk in critical domains such as medicine where people's lives are at stake. Several works have been done to uncover the inner reasoning of deep neural networks. Saliency methods explain model decisions by assigning weights to input features that reflect their contribution to the classifier decision. However, not all features are necessary to explain a model decision. In practice, classifiers might strongly rely on a subset of features that might be sufficient to explain a particular decision. The aim of this article is to propose a method to simplify the prediction explanation of One-Dimensional (1D) Convolutional Neural Networks (CNN) by identifying sufficient and necessary features-sets. We also propose an adaptation of Layer-wise Relevance Propagation for 1D-CNN. Experiments carried out on multiple datasets show that the distribution of relevance among features is similar to that obtained with a well known state of the art model. Moreover, the sufficient and necessary features extracted perceptually appear convincing to humans.
Kameni Florentin Flambeau Jiechieu, Norbert Tsopzé
Neural Computing and Applications
Félicité Gamgne Domgue, Norbert Tsopzé, René Ndoundam
HAL (Le Centre pour la Communication Scientifique Directe)
Community detection in directed networks appears as one of the most relevant topics in the field of network analysis. One common theme in some formalizations is that flows should tend to stay within communities and could be centered round core nodes called "kernels". Hence, we expecttriads to play an important role in the detection of that kind of community. Triads for directed graph are directed sub-graphs of 3 nodes involving at least 2 links between them. To identify communities in directed networks, this paper proposes an undirected edge-weighting scheme based on directed triads. We also propose a new metric on quality of the communities that is based on the triad density of communities. To validate our approach, an experiment was conducted on some networks which show that it has better performance on triad-based density over some state-of-the-art methods.
Norbert Tsopzé, Gamgne Domgue Félicité
HAL (Le Centre pour la Communication Scientifique Directe)
Mickaël Coustaty, Norbert Tsopzé, Alain Bouju, Karell Bertet, Georges Louis
HAL (Le Centre pour la Communication Scientifique Directe)
International audience
Norbert Tsopzé, Clément Guerin, Karell Bertet, Arnaud Revel
HAL (Le Centre pour la Communication Scientifique Directe)
8 pages
Mickaël Coustaty, Alain Bouju, Georges Louis, Norbert Tsopzé, Karell Bertet, Jean-Marc Ogier
HAL (Le Centre pour la Communication Scientifique Directe)
International audience
Mickaël Coustaty, Norbert Tsopzé, Karell Berthet, Alain Bouju, Georges Louis
Les cahiers du numérique
Cet article s'intresse au traitement des documents anciens, et plus particulirement des images de lettrines (lettres dcores en dbut de paragraphe de livres du 16 e sicle). Cette approche s'appuie sur une dmarche ontologique pour tablir les liens entre les rgions extraites par les algorithmes de traitement d'images d'une part et les lments smantiques d'autre part, afin d'aider les historiens dans leur interprtation et les situer dans le temps. Nous avons combin trois ontologies (thsaurus dfini par les historiens, ontologie des traitements informatiques et ontologie spatiale) pour l'annotation de ces images ; ainsi nous avons donc dfini des rgles permettant d'annoter certaines rgions comme tant la lettre, une partie du corps de personnage ou encore de caractriser une lettrine comme figurative.
Norbert Tsopzé, Engelbert Mephu Nguifo, Gilbert Tindo
HAL (Le Centre pour la Communication Scientifique Directe)
International audience
Norbert Tsopzé, Engelbert Mephu Nguifo, Gilbert Tindo
Techniques et sciences informatiques
This paper presents a new method of artificial neural network architecture definition. This approach shows how the concept lattice could help to define the artificial neural network architecture. We will also present the existing approaches that are reported in the litterature. Finally a comparative study is described and discussed.
Les reseaux de neurones artificiels connaissent des succes dans plusieurs domaines. Maisles utilisateurs des reseaux de neurones sont souvent confrontes aux problemes de definitionde son architecture et d’interpretabilite de ses resultats. Plusieurs travaux ont essaye d’apporterune solution a ces problemes. Pour les problemes d’architecture, certains auteurs proposentde deduire cette architecture a partir d’un ensemble de connaissances decrivant le domaine duprobleme et d’autres proposent d’ajouter de maniere incrementale les neurones a un reseauayant une taille initiale minimale. Les solutions proposees pour le probleme d’interpretabilitedes resultats consistent a extraire un ensemble de regles decrivant le fonctionnement du reseau.Cette these contribue a la resolution de ces deux problemes. Nous nous limitons a l’utilisationdes reseaux de neurones dans la resolution des problemes de classification.Nous presentons dans cette these un etat de l’art des methodes existantes de recherche d’architecturede reseaux de neurones : une etude theorique et experimentale est aussi faite. Decette etude, nous observons comme limites de ces methodes la disponibilite absolue des connaissancespour construire un reseau interpretable et la construction des reseaux difficiles a interpreteren absence de connaissances. En alternative, nous proposons une methode appelee CLANN(Concept Lattice-based Artificial Neural network) basee les treillis de Galois qui construit undemi-treillis a partir des donnees et deduire de ce demi-treillis l’architacture du reseau. CLANNetant limitee a la resolution des problemes a deux classes, nous proposons MCLANN permettantd’etendre cette methodes de recherche d’architecture des reseaux de neurones aux problemes aplusieurs classes.Nous proposons aussi une methode appelee ’Approche des MaxSubsets’ pour l’extractiondes regles a partir d’un reseau de neurones. La particularite de cette methode est la possibilited’extraire les deux formats de regles (’si alors’ et ’m parmi N’) a partir d’une structure quenous construisons. Nous proposons aussi une facon d’expliquer le resultat calcule par le reseauconstruit par la methode MCLANN au sujet d’un exemple.
Norbert Tsopzé, Engelbert Mephu Nguifo, G. Tindo
HAL (Le Centre pour la Communication Scientifique Directe)
National audience
Norbert Tsopzé, Engelbert Mephu Nguifo, Gilbert Tindo
EGC eBooks
Norbert Tsopzé, Engelbert Mephu Nguifo, Gilbert Tindo
HAL (Le Centre pour la Communication Scientifique Directe)
National audience
Researcher profile
Scientific publications
2 publications
3 publications
7 publications
2 publications
3 publications
5 publications
7 publications
2 publications
1 publication
1 publication
3 publications
1 publication
3 publications
2 publications
7 publications
2 publications
2 publications
1 publication
1 publication
5 publications
Collaborations