Mohammed Barrouch

Mr. Mohammed Barrouch

orcid

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
27 April 2026 in Symmetry
Verified
Journal Article
Open access
PDF file
BZBilal Ben Zahra
Bilal Ben Zahra
MBMohammed Barrouch
Mohammed Barrouch
CWCharchaoui Wiam
Charchaoui Wiam
AAAbdellah Ahourag
Abdellah Ahourag
EKEl Moutaouakil Karim
El Moutaouakil Karim
Vasile PaladeVasile Palade
Vasile Palade
This paper introduces the Quantum-Inspired Impulsive Continuous Hopfield Network (Q-ICHN), a novel hybrid control framework designed to handle non-smooth, high-energy perturbations in nonlinear dynamical systems. Standard Continuous Hopfield Networks (CHNs) rely on sigmoidal activation functions that are prone to gradient saturation, which leads to an insufficient corrective response when the system undergoes large deviations from equilibrium. To overcome this shortcoming, the proposed Q-ICHN adopts a wave-packet-based activation function grounded in the stationary Schrödinger equation, yielding a non-monotonic and oscillatory activation profile that sustains effective compensatory dynamics across a broad range of states. Furthermore, the proposed framework incorporates Madelung’s quantum potential into the control architecture, thereby enabling a fundamental reshaping of the system’s energy landscape. Specifically, this induces a tunneling-like mechanism that allows the system to circumvent local minima and rapidly recover from impulsive disturbances, manifested as a sharpened attractor structure in the phase-space domain. Together, these properties yield enhanced convergence behavior and improved robustness over traditional neural control approaches. To rigorously assess its merits, the performance of the Q-ICHN is evaluated through a large-scale benchmark involving 20 established control methods, including Sliding Mode Control (SMC), Model Predictive Control (MPC), and Backstepping. The experimental results obtained across 20 heterogeneous scenarios demonstrate that the proposed model achieves a 48% reduction in Mean Squared Error (MSE) relative to the classical ICHN. In addition, the Q-ICHN exhibits improved smoothness, reflected in a 30% reduction in jerk with respect to high-gain robust controllers, and enhanced reliability, validated by superior spectral purity and a 34% reduction in integrated variance under stochastic perturbations. Collectively, these results underscore the potential of quantum-inspired activation mechanisms to favorably balance control responsiveness and harmonic stability, providing a robust framework for handling both continuous dynamics and impulsive effects.

16 November 2026 in Information
Verified
Journal Article
Open access
PDF file
AOAbdelfattah OULAIKA
Abdelfattah OULAIKA
MBMohammed Barrouch
Mohammed Barrouch
ZBZAKARIA BOUHANCH
ZAKARIA BOUHANCH
AAAbdellah Ahourag
Abdellah Ahourag
MBMohamed Barhdadi
Mohamed Barhdadi
Vasile PaladeVasile Palade
Vasile Palade
Agent-based models are widely used to study complex adaptive dynamics in heterogeneous populations, largely because they capture how autonomous agents behave without central coordination. The same decentralization makes stability hard to establish analytically. We take one controlled two-state contagion model on a graph, steered by a single broadcast scalar, and study six controllers on it: a fuzzy inference system, a tabular reinforcement learner, and four ways of combining them. The analysis is a stochastic drift argument on a Perron-weighted Lyapunov function that asks of each controller only a lower bound on its effort above an activation level, not a Lyapunov function of its own. The spectral radius of the interaction operator sets a critical effort, and a controller whose floor exceeds it keeps the population regulated. Each hybrid inherits a floor from the two it combines, which separates the four: safety blending and the hierarchical override keep a floor whatever the learner does, convex fusion and switching do not. Asynchronous updates follow from the same argument; delays and interruptions do not, and are left to simulation. Its application settings are readings of one model rather than independent validations. The trained learner cuts the fuzzy controller’s cost by 61%, yet eight tuned classical and model-based baselines all cost less than any of the six. The hybrids contain the learner’s failure outside its training range, but that failure is produced by the training protocol: a wider range of initial states removes it. A controller call costs the same up to 106 agents, and three further models mark the scope: control is essential in a cascade, marginal under segregation, and counterproductive where it acts non-monotonically.

2 November 2026 in Information
Verified
Journal Article
Open access
PDF file
WCWiam Charchaoui
Wiam Charchaoui
MBMohammed Barrouch
Mohammed Barrouch
AOAzedine Ouhmid
Azedine Ouhmid
MBMohamed Barhdadi
Mohamed Barhdadi
EKEl Moutaouakil Karim
El Moutaouakil Karim
Graph neural networks (GNNs) capture local image structure with few parameters, whereas Vision Transformers (ViTs) model global context at a higher computational cost. We propose the Omni-Flex GraphFormer, which applies a graph branch (OMNIVIG: max-relative aggregation over a fixed 3×3 pixel neighbourhood) and a Transformer branch (FLEXIVIT: global multi-head self-attention) in parallel and sums their outputs. Because each baseline is the hybrid with one branch removed, comparing the three models on eight datasets from 8×8 to 96×96 pixels, under equal training budgets and over three seeds, is a controlled branch ablation. Global attention is decisive: the hybrid exceeds OMNIVIG in test macro-F1 on every dataset and in all 24 runs, by 3.7 to 15.1 points (Wilcoxon test, p=0.0078). Against FLEXIVIT, the hybrid has the best mean F1 on six datasets and a 40–62% smaller seed-to-seed spread on CIFAR-10 and CIFAR-100, but the difference is not significant (p=0.38). The graph branch adds only 0.5–3.1% to the number of multiply–accumulate operations, and under warmed-up timing, both models run at the same speed, whereas unwarmed timing overstates latency by up to 15×. Under Gaussian noise, the summed hybrid degrades like the graph branch, which motivates gated or attention-guiding fusion.