Toumi Bouchentouf
Research Subjects
Toumi Bouchentouf has a total of 59 co-authors. Here are the top co-authors by number of shared publications:
- Jamal Berrich
National School of Applied Sciences, Université Mohammed Premier
17 shared publications - Mohammed Ghaouth Belkasmi
Mohammed First University
11 shared publications - Mohammed Saber
Mohammed First University, ENSAO, SmartICT
8 shared publications Adil Zeaaraoui
Mohammed First University
7 shared publicationsEl Hassane Ettifouri
Mohammed First University, Oujda, Morocco
7 shared publications- Zineb Bougroun
Laboratory of Applied Mathematics, Signal Processing and Computer Science, Department of Computer Science, ESTO, Oujda, Morocco
7 shared publications
Latest Publications
- 2 January 2026
- Published by Springer Nature
- p. 491-501
Transfer Learning in Reinforcement Learning for Mobile Robots: A Comprehensive Overview
- 2 January 2026
- Published by Springer Nature
- p. 491-501
Transfer Learning in Reinforcement Learning for Mobile Robots: A Comprehensive Overview
- 2 April 2025
- Published by Springer Nature
- p. 67-74
Fast Drone Obstacle Detection Approach Based on MobileNet Classification CNN Architectures
- 2 April 2025
- Published by Springer Nature
- p. 67-74
Fast Drone Obstacle Detection Approach Based on MobileNet Classification CNN Architectures
- Open Access
- 7 March 2025
- Published by Qeios Ltd
Review of: "LLM Modules: Knowledge Transfer from a Large to a Small Model Using Enhanced Cross-Attention"
This work makes a significant contribution to the field of knowledge transfer in language models. The modular approach and innovative use of ENHANCED CROSS-ATTENTION provide an effective alternative to classical distillation methods, with clear advan...
- 7 March 2025
- Published by Qeios Ltd
Open AccessReview of: "LLM Modules: Knowledge Transfer from a Large to a Small Model Using Enhanced Cross-Attention"
This work makes a significant contribution to the field of knowledge transfer in language models. The modular approach and innovative use of ENHANCED CROSS-ATTENTION provide an effective alternative to classical distillation methods, with clear advan...
- Open Access
- 31 December 2024
- Published by Politechnika Lubelska in Applied Computer Science
- Vol. 20 (4) , 138-148
EVALUATING LARGE LANGUAGE MODELS FOR MEDICAL INFORMATION EXTRACTION: A COMPARATIVE STUDY OF ZERO-SHOT AND SCHEMA-BASED METHODS
This study investigates the application of large language models, particularly ChatGPT, in the extraction and structuring of medical information from free-text patient reports. The authors explore two distinct methods: a zero-shot extraction approach and a schema-based extraction approach. The dataset, consisting of 1230 anonymized French medical reports from the Department of Neonatology of the Mohammed VI University Hospital, served as the basis for these experiments. The findings indicate that while ChatGPT demonstrates a significant capability in structuring medical data, certain challenges remain, particularly with complex and non-standardized text formats. The authors evaluate the model's performance using precision, recall, and F1 score metrics, providing a comprehensive assessment of its applicability in clinical settings.
- 31 December 2024
- Published by Politechnika Lubelska in Applied Computer Science
- Vol. 20 (4) , 138-148
Open AccessEVALUATING LARGE LANGUAGE MODELS FOR MEDICAL INFORMATION EXTRACTION: A COMPARATIVE STUDY OF ZERO-SHOT AND SCHEMA-BASED METHODS
This study investigates the application of large language models, particularly ChatGPT, in the extraction and structuring of medical information from free-text patient reports. The authors explore two distinct methods: a zero-shot extraction approach and a schema-based extraction approach. The dataset, consisting of 1230 anonymized French medical reports from the Department of Neonatology of the Mohammed VI University Hospital, served as the basis for these experiments. The findings indicate that while ChatGPT demonstrates a significant capability in structuring medical data, certain challenges remain, particularly with complex and non-standardized text formats. The authors evaluate the model's performance using precision, recall, and F1 score metrics, providing a comprehensive assessment of its applicability in clinical settings.
- 29 August 2024
- Published by Springer Nature
- p. 523-532
Deep Reinforcement Learning for Mobile Robots: Overview and Issues
- 29 August 2024
- Published by Springer Nature
- p. 523-532
Deep Reinforcement Learning for Mobile Robots: Overview and Issues
- 29 August 2024
- Published by Springer Nature
- p. 502-513
Unmanned Aerial Vehicles (UAVs): from Evolution to Simulation, A Comprehensive Exploration
- 29 August 2024
- Published by Springer Nature
- p. 502-513
Unmanned Aerial Vehicles (UAVs): from Evolution to Simulation, A Comprehensive Exploration
- 28 June 2023
- Published by Springer Nature in Journal of Real-Time Image Processing
- Vol. 20 (4) , 1-10
Evolutionary algorithm for optimized CNN architecture search applied to real-time boat detection in aerial images
- 28 June 2023
- Published by Springer Nature in Journal of Real-Time Image Processing
- Vol. 20 (4) , 1-10
Evolutionary algorithm for optimized CNN architecture search applied to real-time boat detection in aerial images
- Open Access
- 13 March 2023
- Published by MDPI AG in Information
- Vol. 14 (3) , 179
Adapting Off-the-Shelf Speech Recognition Systems for Novel Words
Current speech recognition systems with fixed vocabularies have difficulties recognizing Out-of-Vocabulary words (OOVs) such as proper nouns and new words. This leads to misunderstandings or even failures in dialog systems. Ensuring effective speech recognition is crucial for the proper functioning of robot assistants. Non-native accents, new vocabulary, and aging voices can cause malfunctions in a speech recognition system. If this task is not executed correctly, the assistant robot will inevitably produce false or random responses. In this paper, we used a statistical approach based on distance algorithms to improve OOV correction. We developed a post-processing algorithm to be combined with a speech recognition model. In this sense, we compared two distance algorithms: Damerau–Levenshtein and Levenshtein distance. We validated the performance of the two distance algorithms in conjunction with five off-the-shelf speech recognition models. Damerau–Levenshtein, as compared to the Levenshtein distance algorithm, succeeded in minimizing the Word Error Rate (WER) when using the MoroccanFrench test set with five speech recognition systems, namely VOSK API, Google API, Wav2vec2.0, SpeechBrain, and Quartznet pre-trained models. Our post-processing method works regardless of the architecture of the speech recognizer, and its results on our MoroccanFrench test set outperformed the five chosen off-the-shelf speech recognizer systems.
- 13 March 2023
- Published by MDPI AG in Information
- Vol. 14 (3) , 179
Open AccessAdapting Off-the-Shelf Speech Recognition Systems for Novel Words
Current speech recognition systems with fixed vocabularies have difficulties recognizing Out-of-Vocabulary words (OOVs) such as proper nouns and new words. This leads to misunderstandings or even failures in dialog systems. Ensuring effective speech recognition is crucial for the proper functioning of robot assistants. Non-native accents, new vocabulary, and aging voices can cause malfunctions in a speech recognition system. If this task is not executed correctly, the assistant robot will inevitably produce false or random responses. In this paper, we used a statistical approach based on distance algorithms to improve OOV correction. We developed a post-processing algorithm to be combined with a speech recognition model. In this sense, we compared two distance algorithms: Damerau–Levenshtein and Levenshtein distance. We validated the performance of the two distance algorithms in conjunction with five off-the-shelf speech recognition models. Damerau–Levenshtein, as compared to the Levenshtein distance algorithm, succeeded in minimizing the Word Error Rate (WER) when using the MoroccanFrench test set with five speech recognition systems, namely VOSK API, Google API, Wav2vec2.0, SpeechBrain, and Quartznet pre-trained models. Our post-processing method works regardless of the architecture of the speech recognizer, and its results on our MoroccanFrench test set outperformed the five chosen off-the-shelf speech recognizer systems.
- Open Access
- 1 December 2022
- Published by Institute of Advanced Engineering and Science in IAES International Journal of Artificial Intelligence (IJ-AI)
- Vol. 11 (4) , 1588-1595
FrBMedQA: the first French biomedical question answering dataset
FrBMedQA is the first French biomedical question answering dataset, containing 41k+ passage-question instances. It was automatically constructed in a cloze-style manner, from biomedical French Wikipedia articles. To test the validity and difficulty of the dataset, we experimented with four statistical baseline models, a biomedical bidirectional encoder representation from transformers (BERT)-based model, and two French BERT-based language model. We also did human evaluation on a subset of the test set. All the three tested models were not able to surpass the best performing baseline model. Human performance at 61.11% is leading the leaderboard with more than 8% from the best performing model. We made available the dataset and the code to reproduce our results.
- 1 December 2022
- Published by Institute of Advanced Engineering and Science in IAES International Journal of Artificial Intelligence (IJ-AI)
- Vol. 11 (4) , 1588-1595
Open AccessFrBMedQA: the first French biomedical question answering dataset
FrBMedQA is the first French biomedical question answering dataset, containing 41k+ passage-question instances. It was automatically constructed in a cloze-style manner, from biomedical French Wikipedia articles. To test the validity and difficulty of the dataset, we experimented with four statistical baseline models, a biomedical bidirectional encoder representation from transformers (BERT)-based model, and two French BERT-based language model. We also did human evaluation on a subset of the test set. All the three tested models were not able to surpass the best performing baseline model. Human performance at 61.11% is leading the leaderboard with more than 8% from the best performing model. We made available the dataset and the code to reproduce our results.
- Open Access
- 1 December 2022
- Published by Institute of Advanced Engineering and Science in IAES International Journal of Artificial Intelligence (IJ-AI)
- Vol. 11 (4) , 1607-1614
A survey and analysis of intrusion detection models based on information security and object technology-cloud intrusion dataset
Nowadays society, economy, and critical infrastructures have become principally dependent on computers, networks, and information technology solutions, on the other side, cyber-attacks are becoming more sophisticated and thus presenting increasing challenges in accurately detecting intrusions. Failure to prevent intrusions could compromise data integrity, confidentiality, and availability. Different detection methods are proposed to tackle computer security threats, which can be broadly classified into anomaly-based intrusion detection systems (AIDS) and signature-based intrusion detection systems (SIDS). One of the most preferred AIDS mechanisms is the machine learning-based approach which provides the most relevant results ever, but it still suffers from disadvantages like unrepresentative dataset, indeed, most of them were collected during a limited period of time, in some specific networks and mostly don't contain up-to-date data. Additionally, they are imbalanced and do not hold sufficient data for all types of attacks, especially new attack types. For this reason, upto-date datasets such as information security and object technology-cloud intrusion dataset (ISOT-CID) are very convenient to train predictive models on a cloud-based intrusion detection approach. The dataset has been collected over a sufficiently long period and involves several hours of attack data, culminating into a few terabytes. It is large and diverse enough to accommodate machine-learning studies.
- 1 December 2022
- Published by Institute of Advanced Engineering and Science in IAES International Journal of Artificial Intelligence (IJ-AI)
- Vol. 11 (4) , 1607-1614
Open AccessA survey and analysis of intrusion detection models based on information security and object technology-cloud intrusion dataset
Nowadays society, economy, and critical infrastructures have become principally dependent on computers, networks, and information technology solutions, on the other side, cyber-attacks are becoming more sophisticated and thus presenting increasing challenges in accurately detecting intrusions. Failure to prevent intrusions could compromise data integrity, confidentiality, and availability. Different detection methods are proposed to tackle computer security threats, which can be broadly classified into anomaly-based intrusion detection systems (AIDS) and signature-based intrusion detection systems (SIDS). One of the most preferred AIDS mechanisms is the machine learning-based approach which provides the most relevant results ever, but it still suffers from disadvantages like unrepresentative dataset, indeed, most of them were collected during a limited period of time, in some specific networks and mostly don't contain up-to-date data. Additionally, they are imbalanced and do not hold sufficient data for all types of attacks, especially new attack types. For this reason, upto-date datasets such as information security and object technology-cloud intrusion dataset (ISOT-CID) are very convenient to train predictive models on a cloud-based intrusion detection approach. The dataset has been collected over a sufficiently long period and involves several hours of attack data, culminating into a few terabytes. It is large and diverse enough to accommodate machine-learning studies.