Abdellah Idrissi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Abdellah Idrissi has a total of 66 co-authors. Here are the top co-authors by number of shared publications:
Manar Abourezq
Faculty of Sciences Mohammed V University In Rabat, Department of Computer Science, 4 Avenue Ibn Batouta, B.P. 1014, Rabat, 10106, Morocco
9 shared publicationsHajar Rehioui
Computer Science Laboratory (LRI), Computer Science Department, Mohammed V University in Rabat
7 shared publications- Sara Retal
(2IACS) Hassan II University Of Casablanca
7 shared publications - Abdelquoddouss Laghrissi
Aalto University, Espoo, Finland
6 shared publications - Fadoua Yakine
Computer Science Laboratory (LRI) Department of Computer Science, Faculty of Sciences, University Mohammed V - Agdal, Rabat
6 shared publications - Youness Boutyour
Department of Computer Science, Faculty of Sciences, Mohammed V University, Rabat, Morocco
6 shared publications
Latest Publications
- 20 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. PP, 1-10
FA-MADT: Enhancing Offline Multi-Agent Reinforcement Learning with Factorized Attention and Decision Transformers
Multi-Agent Reinforcement Learning (MARL) is a challenging issue in respect of scalability, coordination, and stability, particularly in the offline setting where exploration is restricted. Decision Transformers (DT) are an emerging technology in offline Reinforcement Learning (RL) for single agents by transforming reinforcement learning into a sequence modeling problem, but their use in multi-agent environments is not fully explored. In this work, we introduce Factorized Attention for Multi-Agent Decision Transformers (FA-MADT), new architecture that enhances coordination and sample efficiency, with design considerations aimed at improving scalability. FA-MADT uses Factorized Attention (FA) to model inter-agent dependencies and thus avoids the quadratic complexity of standard self-attention while preserving relevant coordination information. With the integration of Return-to-Go (RTG) conditioning, FA-MADT is capable of making trajectory-based decisions and thus performs well in long-term planning without the need for online exploration. Furthermore, Behavior Cloning (BC) regularization improves policy learning by preventing out-of-distribution (OOD) actions and enhancing the generality of the policy over different offline datasets. We evaluate FA-MADT on three benchmark suites—Multi-Agent MuJoCo, the StarCraft Multi-Agent Challenge (SMAC), and multi-agent traffic signal control—demonstrating consistent improvements over state-of-the-art baselines including MADT, TransMix, CQL-MA, and OMIGA. Our method improves coordination efficiency by up to 15%, reduces OOD action rates by 20%, and lowers memory usage by 12%. FA-MADT also reduces attention complexity from $\mathcal{O}(N^{2})$ to $\mathcal{O}(N \cdot d \cdot k)$ with $k \ll N$, supporting scalable policy learning. Additionally, BC regularization improves OOD action selection accuracy by up to 9.4% on the most challenging SMAC scenarios, contributing to more stable offline policy optimization. These results highlight FA-MADT as a promising step toward scalable and generalizable offline multi-agent decision-making, with future work needed to validate its robustness in real-world systems involving noisy sensors and physical dynamics.
- 20 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. PP, 1-10
FA-MADT: Enhancing Offline Multi-Agent Reinforcement Learning with Factorized Attention and Decision Transformers
Multi-Agent Reinforcement Learning (MARL) is a challenging issue in respect of scalability, coordination, and stability, particularly in the offline setting where exploration is restricted. Decision Transformers (DT) are an emerging technology in offline Reinforcement Learning (RL) for single agents by transforming reinforcement learning into a sequence modeling problem, but their use in multi-agent environments is not fully explored. In this work, we introduce Factorized Attention for Multi-Agent Decision Transformers (FA-MADT), new architecture that enhances coordination and sample efficiency, with design considerations aimed at improving scalability. FA-MADT uses Factorized Attention (FA) to model inter-agent dependencies and thus avoids the quadratic complexity of standard self-attention while preserving relevant coordination information. With the integration of Return-to-Go (RTG) conditioning, FA-MADT is capable of making trajectory-based decisions and thus performs well in long-term planning without the need for online exploration. Furthermore, Behavior Cloning (BC) regularization improves policy learning by preventing out-of-distribution (OOD) actions and enhancing the generality of the policy over different offline datasets. We evaluate FA-MADT on three benchmark suites—Multi-Agent MuJoCo, the StarCraft Multi-Agent Challenge (SMAC), and multi-agent traffic signal control—demonstrating consistent improvements over state-of-the-art baselines including MADT, TransMix, CQL-MA, and OMIGA. Our method improves coordination efficiency by up to 15%, reduces OOD action rates by 20%, and lowers memory usage by 12%. FA-MADT also reduces attention complexity from $\mathcal{O}(N^{2})$ to $\mathcal{O}(N \cdot d \cdot k)$ with $k \ll N$, supporting scalable policy learning. Additionally, BC regularization improves OOD action selection accuracy by up to 9.4% on the most challenging SMAC scenarios, contributing to more stable offline policy optimization. These results highlight FA-MADT as a promising step toward scalable and generalizable offline multi-agent decision-making, with future work needed to validate its robustness in real-world systems involving noisy sensors and physical dynamics.
- 28 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cognitive and Developmental Systems
- Vol. 17 (3) , 540-553
Adaptive Environment Generation for Continual Learning: Integrating Constraint Logic Programming With Deep Reinforcement Learning
In this paper, we introduce a novel framework that combines Constraint Logic Programming (CLP) with Deep Reinforcement Learning (DRL) to create adaptive environments for continual learning. We focus on two challenging domains: Sudoku puzzles and scheduling problems, where environment complexity evolves based on the agent’s performance. By integrating CLP, we dynamically adjust problem difficulty in response to the agent’s learning trajectory, ensuring a progressively challenging environment that fosters enhanced problem-solving skills. Empirical results across 500,000 episodes show substantial improvements in solve rates, increasing from 6% to 86% for Sudoku puzzles and 7% to 79% for scheduling problems, alongside significant reductions in the average steps required to solve each problem. The proposed adaptive environment generation demonstrates the potential of CLP in advancing RL agents’ continual learning capabilities by dynamically regulating complexity, thus improving their adaptability and learning efficiency. This framework contributes to the broader fields of reinforcement learning and procedural content generation by introducing an innovative approach to continual adaptation in complex environments.
- 28 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cognitive and Developmental Systems
- Vol. 17 (3) , 540-553
Adaptive Environment Generation for Continual Learning: Integrating Constraint Logic Programming With Deep Reinforcement Learning
In this paper, we introduce a novel framework that combines Constraint Logic Programming (CLP) with Deep Reinforcement Learning (DRL) to create adaptive environments for continual learning. We focus on two challenging domains: Sudoku puzzles and scheduling problems, where environment complexity evolves based on the agent’s performance. By integrating CLP, we dynamically adjust problem difficulty in response to the agent’s learning trajectory, ensuring a progressively challenging environment that fosters enhanced problem-solving skills. Empirical results across 500,000 episodes show substantial improvements in solve rates, increasing from 6% to 86% for Sudoku puzzles and 7% to 79% for scheduling problems, alongside significant reductions in the average steps required to solve each problem. The proposed adaptive environment generation demonstrates the potential of CLP in advancing RL agents’ continual learning capabilities by dynamically regulating complexity, thus improving their adaptability and learning efficiency. This framework contributes to the broader fields of reinforcement learning and procedural content generation by introducing an innovative approach to continual adaptation in complex environments.
- 4 October 2024
- Published by Springer Nature
- p. 539-553
Optimization Strategies in Mobile Edge Computing Through Intelligent Task Offloading
- 4 October 2024
- Published by Springer Nature
- p. 539-553
Optimization Strategies in Mobile Edge Computing Through Intelligent Task Offloading
- 4 October 2024
- Published by Springer Nature
- p. 329-342
AI in Adaptive Learning: Challenges and Opportunities
- 4 October 2024
- Published by Springer Nature
- p. 329-342
AI in Adaptive Learning: Challenges and Opportunities
- 4 October 2024
- Published by Springer Nature
- p. 265-284
Artificial Intelligence and Assessment Generators in Education: A Comprehensive Review
- 4 October 2024
- Published by Springer Nature
- p. 265-284
Artificial Intelligence and Assessment Generators in Education: A Comprehensive Review
- 4 October 2024
- Published by Springer Nature
- p. 65-73
Explainable Multi-agent Network for Multi-classification in Small Tabular Data
- 4 October 2024
- Published by Springer Nature
- p. 65-73
Explainable Multi-agent Network for Multi-classification in Small Tabular Data
- 4 October 2024
- Published by Springer Nature
- p. 3-21
Quantum Denclue Algorithm (QDA) as a New Clustering Approach Within Quantum Machine Learning
- 4 October 2024
- Published by Springer Nature
- p. 3-21
Quantum Denclue Algorithm (QDA) as a New Clustering Approach Within Quantum Machine Learning
- 4 October 2024
- Published by Springer Nature
- p. 391-406
Machine Learning to Predict Railway Infrastructure Defects
- 4 October 2024
- Published by Springer Nature
- p. 391-406
Machine Learning to Predict Railway Infrastructure Defects
- 4 October 2024
- Published by Springer Nature
- p. 501-513
Data Quality Assessment of a Utility Company’s Geographic Information System
- 4 October 2024
- Published by Springer Nature
- p. 501-513
Data Quality Assessment of a Utility Company’s Geographic Information System
- 4 October 2024
- Published by Springer Nature
- p. 23-35
Proposal of Three Algorithms Improving the DENCLUE Algorithm for Data Clustering
- 4 October 2024
- Published by Springer Nature
- p. 23-35
Proposal of Three Algorithms Improving the DENCLUE Algorithm for Data Clustering