Hossam Sharara

Dr. Hossam Sharara

orcid

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

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

Representative Publications

View All Publications
2 March 2024 in Cluster Computing
Verified
Journal Article
Open access
PDF file
YAYoussif Abuzied
Youssif Abuzied
MGMohamed Ghanem
Mohamed Ghanem
FDFadi Dawoud
Fadi Dawoud
HGHabiba Gamal
Habiba Gamal
ESEslam Soliman
Eslam Soliman
TETamer ElBatt
Tamer ElBatt
In this paper we introduce a scalable, privacy-preserving, federated learning framework, coined FLoBC, based on the concept of distributed ledgers underlying blockchains. This is motivated by the rapid growth of data worldwide, especially decentralized data which calls for scalable, decenteralized machine learning models which is capable of preserving the privacy of the data of the participating users. Towards this objective, we first motivate and define the problem scope. We then introduce the proposed FLoBC system architecture hinging on a number of key pillars, namely parallelism, decentralization and node update synchronization. In particular, we examine a number of known node update synchronization policies and examine their performance merits and design trade-offs. Finally, we compare the proposed federated learning system to a centralized learning system baseline to demonstrate its performance merits. Our main finding in this paper is that our proposed decentralized learning framework was able to achieve comparable performance to a classic centralized learning system, while distributing the model training process across multiple nodes without sharing their actual data. This provides a scalable, privacy-preserving solution for training a variety of large machine learning models. Graphical abstract

7 December 2023 in Applied Sciences
Verified
Journal Article
Open access
PDF file
NGNada Ayman GabAllah
Nada Ayman GabAllah
HSHossam Sharara
Hossam Sharara
Ahmed RafeaAhmed Rafea
Ahmed Rafea
The spread of fake news on social media continues to be one of the main challenges facing internet users, prohibiting them from discerning authentic from fabricated pieces of information. Hence, identifying the veracity of the content in social posts becomes an important challenge, especially with more people continuing to use social media as their main channel for news consumption. Although a number of machine learning models were proposed in the literature to tackle this challenge, the majority rely on the textual content of the post to identify its veracity, which poses a limitation to the performance of such models, especially on platforms where the content of the users’ post is limited (e.g., Twitter, where each post is limited to 140 characters). In this paper, we propose a deep-learning approach for tackling the fake news detection problem that incorporates the content of both the social post and the associated news article as well as the context of the social post, coined TChecker. Throughout the experiments, we use the benchmark dataset FakeNewsNet to illustrate that our proposed model (TChecker) is able to achieve higher performance across all metrics against a number of baseline models that utilize the social content only as well as models combining both social and news content.

20 January 2025 in IEEE Access
Verified
Journal Article
Open access
IEIsmael Elsharkawi
Ismael Elsharkawi
HSHossam Sharara
Hossam Sharara
Ahmed RafeaAhmed Rafea
Ahmed Rafea
Image representation in computer vision is a long-standing problem that has a great effect on the performance of a model. The traditional methods rely on treating the image as a grid, then using Convolutional Neural Networks (CNNs) to generate an image representation. Afterwards, Vision Transformers and MLP-Mixers were introduced to represent an image as a sequence of features similar to other sources of data. More recently, Vision Graph Neural Network (ViG) have proposed the treatment of an image as a graph of nodes. Nevertheless, the graph construction in ViG depends on k-nearest neighbours (k-nn). This leads to a number of challenges; first, determining the right value for the k in every layer, and more importantly, that this value is the same for all nodes, which leads to limiting the graph expressiveness in capturing the image details, as some parts of the image can be similar to a larger number of other parts, while for other parts, we might be forcing completely irrelevant neighbours to satisfy the k-nn condition. In this work, we propose a new approach that relies on the similarity score thresholding to create the graph edges and, subsequently, the neighboring nodes. Rather than the number of neighbours, we allow the specification of the normalized similarity threshold as an input parameter for each layer, which is more intuitive. We also propose a decreasing threshold framework for selecting the input threshold for all layers. We show that our proposed method is able to achieve higher performance than the ViG model for image classification on the benchmark ImageNet-1K dataset, without increasing the complexity of the model.

1 August 2021
Verified
Conference Paper
Closed access
NGNada Ayman GabAllah
Nada Ayman GabAllah
IFIbrahim Farrag
Ibrahim Farrag
ONOmar El Nawawy
Omar El Nawawy
RKRamy Khalil
Ramy Khalil
HSHossam Sharara
Hossam Sharara
TETamer ElBatt
Tamer ElBatt
In this paper, we design, prototype and evaluate the performance of a novel IoT system with an intelligent edge. Our proposed design spans a modular three-tier IoT system architecture including the edge, gateway and cloud tiers. The proposed system leverages the local data acquired at the edge to build a distributed machine learning framework for local predictions and real-time responses at the edge. In addition, it proposes a more comprehensive model at the gateway for more centralized decision making, leveraging the data collected from multiple edges devices. One of the novel features of our system is minimizing the data transfer from the edge to the gateway tier. This is attained through intelligent data filtering at the edge to focus on the key events that provide the highest information gain to the gateway machine learning model. This, in turn, reduces the amount of noise that can impact the gateway model as well as minimizes the inter-tier communications overhead. To test our proposed system, we implement a prototype spanning the first two-tiers, and show through experiments how our system is capable of achieving a comparable performance to traditional centralized machine learning frameworks, while reducing the inter-tier communications overhead by up to 50%.

12 September 2022
Verified
Conference Paper
Closed access
MIMohamed Tarek Ismail
Mohamed Tarek Ismail
HSHossam Sharara
Hossam Sharara
KMKareem Madkour
Kareem Madkour
KSKarim Seddik
Karim Seddik
Technology scaling has increased the complexity of integrated circuit design. It has also led to more challenges in the field of Design for Manufacturing (DFM). One of these challenges is lithography hotspot detection. Hotspots (HS) are design patterns that negatively affect the output yield. Identifying these patterns early in the design phase is crucial for high yield fabrication. Machine Learning-based (ML) hotspot detection techniques are promising since they have shown superior results to other methods such as pattern matching. Training ML models is a challenging task due three main reasons. First, industrial training designs contain millions of unique patterns. It is impractical to train models using this large number of patterns due to limited computational and memory resources. Second, the HS detection problem has an imbalanced nature; datasets typically have a limited number of HS and a large number of non-hotspots. Lastly, hotspot and non-hotspot patterns can have very similar geometries causing models to be susceptible to high false positive rates. Due to these reasons, the use of data sampling techniques is needed to choose the best representative dataset for training. In this paper, a dataset sampling technique based on autoencoders is introduced. The autoencoders are used to identify latent data features that can reconstruct the input patterns. These features are used to group the patterns using Density-based spatial clustering of applications with noise (DBSCAN). Then, the clustered patterns are sampled to reduce the training set size. Experiments on the ICCAD2019 dataset show that the proposed data sampling approach can reduce the dataset size while maintaining the levels of recall and precision that were obtained using the full dataset.

6 November 2023 in Research Square
Verified
Preprint
Open access
PDF file
YAYoussif Abuzied
Youssif Abuzied
MGMohamed Ghanem
Mohamed Ghanem
FDFadi Dawoud
Fadi Dawoud
HGHabiba Gamal
Habiba Gamal
ESEslam Soliman
Eslam Soliman
TETamer Elbatt
Tamer Elbatt
In this paper we introduce a scalable, privacy-preserving, federated learningframework, coined FLoBC, based on the concept of distributed ledgers underlyingblockchains. This is motivated by the rapid growth of data worldwide, especiallydecentralized data which calls for scalable, decenteralized machine learning mod-els which is capable of preserving the privacy of the data of the participatingusers. Towards this objective, we first motivate and define the problem scope. Wethen introduce the proposed FLoBC system architecture hinging on a number ofkey pillars, namely parallelism, decentralization and node update synchroniza-tion. In particular, we examine a number of known node update synchronizationpolicies and examine their performance merits and design trade-offs. Finally, wecompare the proposed federated learning system to a centralized learning systembaseline to demonstrate its performance merits. Our main finding in this paperis that our proposed decentralized learning framework was able to achieve com-parable performance to a classic centralized learning system, while distributingthe model training process across multiple nodes without sharing their actualdata. This provides a scalable, privacy-preserving solution for training a varietyof large machine learning models.