 |
 |
Volume 42 - Numéro spécial CRI 2023 - 2024
Le Colloque de Recherche en Informatique est la plus importance conférence scientifique en Informatique au Cameroun depuis sa création en 2013. Il regroupe tous les deux ans plus de deux cents chercheurs, enseignants-chercheurs et professionnels de l’Informatique au Cameroun pour discuter des enjeux, défis, opportunités, risques et présenter les innovations et résultats scientifiques dans ce domaine. La sixième édition du Colloque de Recherche en Informatique CRI’2023 (http://cri-info.cm) s’est tenu les 12 et 13 décembre 2023 au Département d’Informatique de la Faculté des Sciences de l’Université de Yaoundé I en traitant les problématiques d’apprentissage artificielle, de fouille de données, d’optimisation combinatoire et du traitement automatique du langage naturel et de la parole.
1. Deux Schémas de Stéganographie Textuelles basés sur le codage des couleurs.
Text steganography is a mechanism of hiding secret text message inside another text as a covering message. In this paper, we propose a text steganographic scheme based on color coding. This includes two different methods: the first based on permutation, and the second based on numeration systems. Given a secret message and a cover text, the proposed schemes embed the secret message in the cover text by making it colored. The stego-text is then send to the receiver by mail. After experiments, the results obtained show that our models perform a better hiding process in terms of hiding capacity as compared to the scheme of Aruna Malik et al. on which our idea is based.
2. Un nouvel algorithme hybride basé sur l'optimisation des colonies de fourmis et les réseaux neuronaux récurrents avec un mécanisme d'attention pour résoudre le problème du voyageur de commerce
In this paper, we propose a hybrid approach for solving the symmetric traveling salesman problem. The proposed approach combines the ant colony algorithm (ACO) with neural networks based on the attention mechanism. The idea is to use the predictive capacity of neural networks to guide the behaviour of ants in choosing the next cities to visit and to use the prediction results of the latter to update the pheromone matrix, thereby improving the quality of the solutions obtained. In concrete terms, attention is focused on the most promising cities by taking into account both distance and pheromone information thanks to the attention mechanism, which makes it possible to assign weights to each city according to its degree of relevance. These weights are then used to predict the next towns to visit for each city. Experimental results on instancesTSP from the TSPLIB library demonstrate that this hybrid approach is better compared to the classic ACO.
3. Analyse des toux COVID-19 : De la forme la plus légère à la plus sévère, une classification réaliste par apprentissage profond
Cough is the most common symptom of lung disease. COVID-19, a respiratory illness, has caused over 700 million positive cases and 7 million deaths worldwide. An effective, affordable, and widely available diagnostic tool is crucial in combating lung disease and the COVID-19 pandemic. Deep learning and machine learning algorithms could be used to analyze the cough sounds of infected patients and make predictions. Our research lab and the COUGHVID research lab provide the cough data. This diagnostic approach can distinguish cough sounds from COVID-19 patients and people suffering from other ailments as well as healthy people using deep learning and feature extraction from Mel spectrograms. The model used is a variant of ConvNet. This ConvNet model can easily capture features in MFCC Vectors and enable convolution parallelism, which increases processing speed. ConvNet attains translational invariance in features through the sharing of weights between layers. During data acquisition for model training, it is important to consider quiet environments to reduce errors in audio quality. The architecture of the convolutional neural networks gives an F1-score of 89%, an accuracy of 90.33% and sensitivity of 87.3%. This system has the potential to significantly impact society by reducing virus transmission, expediting patient treatment, and freeing up hospital resources. Early detection of COVID-19 can prevent disease progression and enhance screening effectiveness.
4. Parallelization of Recurrent Neural Network training algorithm with implicit aggregation on multi-core architectures
Recent work has shown that deep learning algorithms are efficient for various tasks, whether in Natural Language Processing (NLP) or in Computer Vision (CV). One of the particularities of these algorithms is that they are so efficient as the amount of data used is large. However, sequential execution of these algorithms on large amounts of data can take a very long time. In this paper, we consider the problem of training Recurrent Neural Network (RNN) for hate (aggressive) messages detection task. We first compared the sequential execution of three variants of RNN, we have shown that Long Short Time Memory (LSTM) provides better metric performance, but implies more important execution time in comparison with Gated Recurrent Unit (GRU) and standard RNN. To have both good metric performance and reduced execution time, we proceeded to a parallel implementation of the training algorithms. We proposed a parallel algorithm based on an implicit aggregation strategy in comparison to the existing approach which is based on a strategy with an aggregation function. We have shown that the convergence of this proposed parallel algorithm is close to that of the sequential algorithm. The experimental results on an 32-core machine at 1.5 GHz and 62 Go of RAM show that better results are obtained with the parallelization strategy that we proposed. For example, with an LSTM on a dataset having more than 100k comments, we obtained an f-measure of 0.922 and a speedup of 7 with our approach, […]
5. Apprentissage auto-supervisé et multilingue appliqué au Wolof, au Swahili et au Fongbe
Under-resourced languages encounter substantial obstacles in speech recognition owing to the scarcity of resources and limited data availability, which impedes their development and widespread adoption. This paper presents a representation learning model that leverages existing frameworks based on self-supervised learning techniques—specifically, Contrastive Predictive Coding (CPC), wav2vec, and a bidirectional variant of CPC—by integrating them with multilingual learning approaches. We apply this model to three African languages: Wolof, Swahili, and Fongbe. Our evaluation of the resulting representations in a downstream task, automatic speech recognition, utilizing an architecture analogous to DeepSpeech, reveals the model’s capacity to discern language specific linguistic features. The results demonstrate promising performance, achieving Word Error Rates (WER) of 61% for Fongbe, 72% for Wolof, and 88% for Swahili. These findings underscore the potential of our approach in advancing speech recognition capabilities for under-resourced languages, particularly within the African linguistic landscape.
6. Application du modèle de représentation acoustique multilingue XLSR pour la transcription de l'Ewondo
Recently popularized self-supervised models appear as a solution to the problem of low data availability via parsimonious learning transfer. We investigate the effectiveness of these multilingual acoustic models, in this case wav2vec 2.0 XLSR-53 and wav2vec 2.0 XLSR-128, for the transcription task of the Ewondo language (spoken in Cameroon). The experiments were conducted on 11 minutes of speech constructed from 103 read sentences. Despite a strong generalization capacity of multilingual acoustic model, preliminary results show that the distance between XLSR embedded languages (English, French, Spanish, German, Mandarin, . . . ) and Ewondo strongly impacts the performance of the transcription model. The highest performances obtained are around 69% on the WER and 28.1% on the CER. An analysis of these preliminary results is carried out andthen interpreted; in order to ultimately propose effective ways of improvement.