Jinan Charafeddine

@article{dornaika_3401,

title = {Semi-supervised learning for multi-view and non-graph data using Graph Convolutional Networks},

author = {Fadi Dornaika and Jinan Charafeddine and Huaiyuan Xiao and Jingjun Bi},

url = {https://www.sciencedirect.com/science/article/abs/pii/S0893608025000978?via%3Dihub},

year  = {2025},

date = {2025-05-01},

journal = {Neural Networks},

volume = {185},

pages = {107218},

abstract = {Semi-supervised learning with a graph-based approach has become increasingly popular in machine learning, particularly when dealing with situations where labeling data is a costly process. Graph Convolution Networks (GCNs) have been widely employed in semi-supervised learning, primarily on graph-structured data like citations and social networks. However, there exists a significant gap in applying these methods to non-graph multi-view data, such as collections of images. To bridge this gap, we introduce a novel deep semi-supervised multi-view classification model tailored specifically for non-graph data. This model independently reconstructs individual graphs using a powerful semi-supervised approach and subsequently merges them adaptively into a unified consensus graph. The consensus graph feeds into a unified GCN framework incorporating a label smoothing constraint.

To assess the efficacy of the proposed model, experiments were conducted across seven multi-view image datasets. Results demonstrate that this model excels in both the graph generation and semi-supervised classification phases, consistently outperforming classical GCNs and other existing semi-supervised multi-view classification approaches. 1},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{dornaika_3269,

title = {Unified Multi-view Data Clustering: Simultaneous Learning of Consensus Coefficient Matrix and Similarity Graph},

author = {Fadi Dornaika and Sally El Hajjar and Jinan Charafeddine and Nagore Barrena},

url = {https://link.springer.com/article/10.1007/s12559-024-10392-z},

year  = {2025},

date = {2025-01-01},

journal = {Cognitive Computation},

volume = {17},

number = {38},

abstract = {Integrating data from multiple sources or views has become increasingly common in data analysis, particularly in fields

like healthcare, finance, and social sciences. However, clustering such multi-view data poses unique challenges due to the

heterogeneity and complexity of the data sources. Traditional clustering methods are often unable to effectively leverage the

information from different views, leading to suboptimal clustering results. To address this challenge, multi-view clustering

techniques have been developed, aiming to integrate information from multiple views to improve clustering performance.

These techniques typically involve learning a similarity matrix for each view and then combining these matrices to form

a consensus similarity matrix, which is subsequently used for clustering. However, existing approaches often suffer from

limitations such as the need for manual tuning of parameters and the inability to effectively capture the underlying structure

of the data. In this paper, we propose a novel approach for multi-view clustering that addresses these limitations by jointly

learning the consensus coefficient matrix and similarity graph. Unlike existing methods that follow a sequential approach

of first learning the coefficient matrix and then constructing the similarity graph, our approach simultaneously learns both

matrices, ensuring a more regularized consensus graph. Additionally, our method automatically adjusts the weight of each

view, eliminating the need for manual parameter tuning. Our approach involves several key steps. First, we formulate an

optimization problem that jointly optimizes the consensus coefficient matrix, unified spectral projection matrix, coefficient

matrix, and soft cluster assignment matrix. We then propose an efficient algorithm to solve this optimization problem, which

involves iteratively updating the matrices until convergence. To learn the consensus coefficient matrix and similarity graph, we

leverage techniques from matrix factorization and graph-based learning. Specifically, we use a self-representation technique

to learn the coefficient matrix (regularization graPh) and a graph regularization technique to learn the similarity graph. By

jointly optimizing these matrices, we ensure that the resulting consensus graph is more regularized and better captures the

underlying structure of the data. We evaluate our approach on several public image datasets, comparing it against state-of-the-

art multi-view clustering methods. Our experimental results demonstrate that our approach consistently outperforms existing

methods in terms of clustering accuracy and robustness. Additionally, we conduct sensitivity analysis to evaluate the impact

of different hyperparameters on the clustering performance. We present a novel approach for multi-view data clustering

that jointly learns the consensus coefficient matrix and similarity graph. By simultaneously optimizing these matrices, our

approach achieves better clustering performance compared to existing methods. Our results demonstrate the effectiveness and

robustness of our approach across different datasets, highlighting its potential for real-world applications in various domain.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{abou_ali_3271,

title = {Enhancing Generalization and Mitigating Overfitting in Deep Learning for Brain Cancer Diagnosis from MRI},

author = {Mohamad Abou Ali and Jinan Charafeddine and Fadi Dornaika and Ignacio Arganda?Carreras},

url = {https://link.springer.com/article/10.1007/s00723-024-01743-y},

year  = {2025},

date = {2025-01-01},

journal = {Applied Magnetic Resonance},

abstract = {Brain cancer represents a significant global health challenge with increasing inci-

dence and mortality rates. Magnetic Resonance Imaging (MRI) plays a pivotal role

in early detection and treatment planning. This study adopts a systematic approach

across four phases: (1) Optimal Model Selection using the Adam optimizer, empha-

sizing accuracy metrics, weight computation, early stopping, and ReduceLROn-

Plateau techniques. (2) Real-world Scenario Simulation through synthetic per-

turbed datasets created by applying noise, blur (to simulate various magnetic field

strengths: 1T, 1.5T, 3T), and patient motion artifacts (mimicking MRI scanning

motion effects) to the testing data from the BT-MRI dataset, an online published

brain tumor MRI dataset. (3) Optimization involving a range of optimizers (Adam,

Adagrad, Nadam, RMSprop, SGD) and online augmentation techniques (AutoMix,

CutMix, LGCOAMix, PatchUp). (4) Solution Exploration integrating Gaussian

Noise and Blur as augmentation strategies during training to enhance model gener-

alization under diverse conditions. Initial evaluations achieved strong performance,

consistently reaching 99.45% accuracy on the BT-MRI dataset. However, testing

against synthetic perturbed datasets mimicking real-world conditions revealed chal-

lenges in maintaining robust model performance. Despite employing diverse opti-

mization methods and advanced augmentation techniques, this study identifies per-

sistent challenges in ensuring model robustness with synthetic perturbed datasets.

Notably, the integration of Gaussian Noise and Blur during training significantly

improved model resilience. This research underscores the critical role of method-

ological rigor and innovative augmentation strategies in advancing deep learning

applications for precise brain cancer diagnosis using MRI.},

keywords = {},

pubstate = {online},

tppubtype = {article}

}

@article{sun_3024,

title = {LCAMix: Local-and-contour aware grid mixing based data augmentation for medical image segmentation},

author = {Danyang Sun and Fadi Dornaika and Jinan Charafeddine},

url = {https://pdf.sciencedirectassets.com/272144/1-s2.0-S1566253524X00068/1-s2.0-S1566253524002628/main.pdf?X-Amz-Security-Token=IQoJb3JpZ2luX2VjEBcaCXVzLWVhc3QtMSJHMEUCIB%2Bm2yVgKP%2BTa1w1tSg%2FydrF%2Fvb0NtGUi2OtE0dCfm5ZAiEAvS5zznXIlJGGqU0uDk6urwemR%2FjBXI8flOEVsv%2Bqi7MqvAUIj%2F%2F%2F%2F%2F%2F%2F%2F%2F%2F%2FARAFGgwwNTkwMDM1NDY4NjUiDGlNeMR%2BKpLBIx%2FJkiqQBUxi37rUay01aPHdPyUjlEmZScIOm09oNOKMm2j4kAJ%2BBXu4xZ6ts%2FYVWeKloZ9bKO688mVLpZYW96XCoW8DS9E%2FGvhCoLNFgFEV35VB1ai8O7M21%2BlU%2Br8%2FcPbj77ypLNC%2BKDWJV0NPrB1DKOvao6ZpbspbrdK4kPspEAzZ%2FVkFT%2FKiopgMXf6iniq2tANtPaizXkk%2BZaNSAGIFt9OmCXMk6H7VYJALKEF4ZJm9kBslRj5nE5Y6edtx2JSSptvlujKMXhJbwjIiUiX7Ui1B4pUWDVtzmLpVcElBM78VwS0H8TKlHXBl5pyVDCym%2Fp2r%2BZbZdJWW6ZWgrqb26Dqc%2F22RSCtts8Lm%2Fs8vSHEc2cfI2mYpPSj5gL51yaDqoZyA9JcEBsQRAecvv%2F8RkbvtzMzuNBRjn4V8HoQR137Q2pYM%2FJgAly5tkgk5KS8JB2jb%2BaTA88ZJLx6YSDeqJdX2PXwyB1XvrrRy8oUgMmff8s2zVgT1f1dX5iPmZ1jtYEVu40cdvJ88ac8j4zAfWQPmBb%2FvfXWd%2BDVOp7h8ftVwQnI7bLDlFBXbhH3oVb%2FFpJ3Bw3DFakM7j%2FSOVug7DP8dC9h8Dj2oIOW8C4QAgdCeHaDRMfh2F7VZjw3FZf3i631%2BAdBp8r4d4XkYeREmbmiovEoVWBBHoYD6RTIK9lp37AmDnhFSAMr2LEmzG2NypUQ3Hwb7zLu4thOxdGz5MBApBAh44ukY8YuN1ufa38eyKRtCUO%2Fxd2kuIIJ2LVXa3bQYug5Lx%2B9KIKhgqxCrwptvf8OaxtQ5FoYhuqNU0pL1Chohou2d1vld8RViOconvLl0VWMsZ5YyP9ZxaM0xtzdz9jZQUUyYOV92u8ehS7ie1qwrMNDAwrIGOrEBxpUgPvpVhdMKearFfabKLjWh5dtniMHIFTE5cSfjDp3cUuAHaEMlV2m22S1E7%2B%2BqLhJRHUxrx0Ahj9AgijOS6Ej7rGG3xGTKGOnhYGjvrjhMzhpYH7bBR2UDVY%2Ff5JnU3IqbM8%2BmzPkwAtPDE4f74N3zJ3CUErPd4nF0UOXyQABK0JE3XGOZmeqIIDshevKKQuAQWWCmCYyGaPWgjtPkrFNgDEqhkhXN4QWxs%2FJfC2Tb&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Date=20240524T152432Z&X-Amz-SignedHeaders=host&X-Amz-Expires=299&X-Amz-Credential=ASIAQ3PHCVTY4RJ4ZEGT%2F20240524%2Fus-east-1%2Fs3%2Faws4_request&X-Amz-Signature=49fff6e8e4aa2160d63ba58a605230103a570a99a899c33049a2447724567bf6&hash=cf866cff1fcd93c461170f4149fdcc586b3e8c5bd13fbef0361db16227b04cc0&host=68042c943591013ac2b2430a89b270f6af2c76d8dfd086a07176afe7c76c2c61&pii=S1566253524002628&tid=spdf-53f47d50-21b1-4fa1-a93b-92addbddb736&sid=5b8e3010830f534aec5849b2797ace9af2f9gxrqb&type=client&tsoh=d3d3LnNjaWVuY2VkaXJlY3QuY29t&ua=001156565a0352015207&rr=888e4b0cfc530203&cc=fr},

year  = {2024},

date = {2024-10-01},

journal = {Information Fusion},

volume = {110},

pages = {102484},

abstract = {Medical image segmentation often faces challenges related to overfitting, primarily due to the limited and complex training samples. This challenge often prompts the use of self-supervised learning and data augmentation. However, self-supervised learning requires well-defined hand-crafted tasks and multiple training stages. On the other hand, basic image augmentation techniques like cropping, rotation, and flipping, effective for natural scene images, have limited efficacy for medical images due to their isotropic nature.



While regional dropout regularization data augmentation methods have proven effective in image recognition tasks, their application in image segmentation is not as extensively studied. Additionally, existing augmentation methods often operate on square regions, leading to the loss of crucial contour information. This is particularly problematic for medical image segmentation tasks dealing with regions of interest characterized by intricate shapes. In this work, we introduce LCAMix, a novel data augmentation approach designed for medical image segmentation. LCAMix operates by blending two images and their segmentation masks based on their superpixels, incorporating a local-and-contour-aware strategy. The training process on augmented images adopts two auxiliary pretext tasks: firstly, classifying local superpixels in augmented images using an adaptive focal margin, leveraging segmentation ground truth masks as prior knowledge; secondly, reconstructing the two source images using mixed superpixels as mutual masks, emphasizing spatial sensitivity. Our method stands out as a simple, one-stage, model-agnostic, and plug-and-play data augmentation solution applicable to various segmentation tasks. Notably, it requires no external data or additional models. Extensive experiments validate its superior performance across diverse medical segmentation datasets and tasks. The source codes are available at https://github.com/DanielaPlusPlus/DataAug4Medical.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{alirezazadeh_3188,

title = {Mises-Fisher similarity-based boosted additive angular margin loss for breast cancer classification},

author = {Pandar Alirezazadeh and Fadi Dornaika and Jinan Charafeddine},

url = {https://link.springer.com/article/10.1007/s10462-024-10963-4#citeas},

year  = {2024},

date = {2024-10-01},

journal = {Artificial Intelligence Review},

volume = {57},

pages = {326},

abstract = {To enhance the accuracy of breast cancer diagnosis, current practices rely on biopsies and microscopic examinations. However, this approach is known for being time-consuming, tedious, and costly. While convolutional neural networks (CNNs) have shown promise for their efficiency and high accuracy, training them effectively becomes challenging in real-world learning scenarios such as class imbalance, small-scale datasets, and label noises. Angular margin-based softmax losses, which concentrate on the angle between features and classifiers embedded in cosine similarity at the classification layer, aim to regulate feature representation learning. Nevertheless, the cosine similarity's lack of a heavy tail impedes its ability to compactly regulate intra-class feature distribution, limiting generalization performance. Moreover, these losses are constrained to target classes when margin penalties are applied, which may not always optimize effectiveness. Addressing these hurdles, we introduce an innovative approach termed MF-BAM (Mises-Fisher Similarity-based Boosted Additive Angular Margin Loss), which extends beyond traditional cosine similarity and is anchored in the von Mises-Fisher distribution. MF-BAM not only penalizes the angle between deep features and their corresponding target class weights but also considers angles between deep features and weights associated with non-target classes. Through extensive experimentation on the BreaKHis dataset, MF-BAM achieves outstanding accuracies of 99.92%, 99.96%, 100.00%, and 98.05% for magnification levels of ×40, ×100, ×200, and ×400, respectively. Furthermore, additional experiments conducted on the BACH dataset for breast cancer classification, as well as on the LFW and YTF datasets for face recognition, affirm the generalization capability of our proposed loss function.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{shakir_3063,

title = {Novel Approximate Distribution of the Generalized Turbulence Channels for MIMO FSO Communications},

author = {Wafaa Shakir and Ali Mahdi and Hani Hamdan and Jinan Charafeddine and Haitham Satai and Radouane Akrache and Samir Hadad and Jinane Sayah},

url = {https://ieeexplore.ieee.org/document/10568928},

year  = {2024},

date = {2024-08-01},

journal = {Ieee Photonics Journal},

volume = {16},

number = {4},

pages = {1 - 15},

abstract = {In this paper, we develop an innovative series representation for the sum of Rician non-zero boresight pointing error random variates based on the k - ? distribution, which is suitable for multiple-input multiple-output (MIMO) transmission for the first time. Then, using this new representation, we introduce a novel closed-form probability density function (PDF) approximation for the sum of Gamma-Gamma random variates with generalized pointing errors and atmospheric attenuation of MIMO free-space optical (FSO) communications. Statistical Kolmogorov-Smirnov tests confirm the accuracy of this approximation over a wide range of channel conditions. The significance of this approximation is emphasized by deriving closed-form expressions for the ergodic capacity, outage probability, and average bit error rate (BER) using Meijer's G-function. This paper provides a comprehensive analysis of the performance of MIMO FSO systems utilizing the equal gain combining (EGC) diversity technique under various conditions, such as different numbers of transmitter and receiver, turbulence intensities, the effects of non-zero boresight pointing errors, and path attenuation. The results show that using MIMO technology with more transmitters and receivers significantly improves the performance of FSO communication compared to other diversity techniques, including single input single output (SISO), and multiple input single output (MISO) systems. Detailed evaluations of the ergodic capacity, outage probability, and average BER performance at high signal-to-noise ratios provide additional insights. Monte-Carlo simulation results demonstrate the accuracy of the proposed approach.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{dornaika_2943,

title = {Towards unsupervised radiograph clustering for COVID-19: The use of graph-based multi-view clustering},

author = {Fadi Dornaika and Sally El Hajjar and Jinan Charafeddine},

url = {https://www.sciencedirect.com/science/article/pii/S0952197624004949?via%3Dihub},

year  = {2024},

date = {2024-07-01},

journal = {Engineering Applications Of Artificial Intelligence},

volume = {133},

number = {Part D},

pages = {108336},

abstract = {Automatic classification methods widely used for diagnosing and analyzing COVID-19 cases. These methods assume known labels and rely on a single view of the dataset. Given the prevalence of COVID-19 cases and the extensive volume of patient records lacking labels, this communication underscores our unique approach?conducting the first study on COVID-19 case diagnosis in an unsupervised manner. Our work operates under the assumption of prior knowledge regarding the number of classes, such as COVID-19, pneumonia, and normal, in a case study.



By adopting an unsupervised learning paradigm, we leverage the wealth of unlabeled data, reducing dependence on human experts for annotating numerous images. This paper introduces an enhanced version of a recent direct method where non-negative cluster indices and spectral embeddings are jointly estimated. Beyond the inherent advantages of this method, our proposed model introduces improvements through two additional types of constraints: (i) ensuring consistent smoothing of cluster labels across all views and (ii) imposing an orthogonality constraint on the matrix of cluster assignments. The efficacy of the proposed method is demonstrated using the public COVIDx dataset with three classes, showcasing promising results in categorizing radiographs. The proposed approach is tested on other public image datasets to assess its effectiveness.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}