Computer

Permanent URI for this collectionhttps://dspace.univ-boumerdes.dz/handle/123456789/3082

Browse

Search Results

Now showing 1 - 10 of 10

Design and implementation of spiking neural networks on FPGA for event-based spatio-temporal applications.
(Université M’hamed Bougara de Boumerdes : Institut de Genie Electrique et Electronique, 2024) Boumerzoug, Nadhir; Zerrari, Dhia Elhak; Cherifi, Dalila (Supervisor)
Inspired by the intricacies of real biological neural systems, Spiking Neural Networks (SNNs) represent an advanced type of artificia lneura lnetwork .SNN soperat ewith discrete spikes, closely mimicking the way neurons communicate in the human brain. This unique method of information processing not only enhances the computational efficien cy ofSN Nsb utal soope ns upn ewpossibiliti esf ordevelopi nglow-pow erneural network systems. In this work, we proposed a generic hardware design of an SNN based on Field-Programmable Gate Arrays (FPGA). The proposed design was implemented and tested with the event-based benchmark dataset “Neuromorphic-MNIST” and managed to achieve a low power consumption and latency, while requiring very minimal hardware resources, all this for an evaluated accuracy.
Gas turbine trip prediction with time-Series data using RNN and LSTM.
(Université M’hamed Bougara de Boumerdes : Institut de Genie Electrique et Electronique, 2024) Bayou, Meroua; Cherifi, Dalila (Supervisor)
Gas turbine (GT) trip is one of the most disruptive occurrences that influenc eG Toperation, as it reduces the remaining useful life of the equipment and results in revenue loss due to business interruption. Thus, early diagnosis of early GT trip symptoms is critical for ensuring effective operation and lowering operating and maintenance expenses. In this work, we implement two neural network methods, RNN and LSTM, for gas turbine trip prediction using a time series sensors readings dataset and compare their performances in accurately predicting gas turbine trips within 60 seconds of their occurrence, allowing operators to take timely and effective actions to prevent trips and ensure the reliability and efficienc yo fpowe rgeneration systems. The objective of this work is to defin eth ebes tperformin gmode lfo rthi ssensitiv etas kand reach the highest possible accuracy and precision by implementing different architectures and exploring variations of hyperparameters such as the number of features, validation split, and the input sequence length. Our experimental results show that both RNN and LSTM are effective in achieving the goal of predicting gas turbine trips prior to their occurrence. The best-performing model is the bidirectional LSTM with multiple features input, a sequence length of 50, and 10% validation split, where we reached a test accuracy of 96.47%, precision 97.14%, recall 94.44%, F1 score 95.77%, and ROC-AUC of 0.96.
Gastrointestinal diseases diagnosis using capsule endoscopy and YOLOv8.
(Université M’hamed Bougara de Boumerdes : Institut de Genie Electrique et Electronique, 2024) Sad Saoud, Abdeldjalil; Korichi, Aymen; Cherifi, Dalila (Supervisor)
Gastrointestinal (GI) diseases represent a substantial global health concern that annually affect millions of individuals and result in nearly two million deaths, highlighting the urgent need for early and accurate diagnosis, as undiagnosed cases can be life-threatening. Wireless Capsule Endoscopy (WCE) is a cutting-edge technology that enables the visualization of gastrointestinal diseases. By capturing thousands of frames per patient, it reduces the risk of human error and increases the accuracy of diagnoses. The enormous volume of images generated by WCE poses a significant challenge for manual diagnosis, prompting the development of computer-aided techniques to enhance the diagnostic process with high accuracy and within a short period. Moreover, deep learning algorithms have demonstrated remarkable performance in medical imaging tasks and especially in GI diseases classification, leveragin gth evast amounts of data generated by WCE to improve diagnostic precision and enhance patient outcomes. This project represents a technique aimed at developing a robust YOLOv8-cls model for gastrointestinal disease classification. By training the model on the Kvasir dataset, the backbone of YOLOv8-cls learns to capture robust and informative features from the input images. These features serve as a powerful representation of the image content. Finally, the extracted features are fed into a classification head, whic hi sfine-tuned to predict the class of the input WCE image, enabling accurate diagnosis of gastrointestinal diseases. In our project we conducted a serie of four experiments to develop a high-performing YOLOv8-cls model for gastrointestinal disease classification. The initial experiment identified the best-performin gYOLOv8-cl svariant, which was then optimized usin ghyperparameter tuning. The final model wa sconstructed using the mixture of experts technique, combining the best-performing variant with optimized hyperparameters. The top-performing variant from experiment 1 achieved an accuracy of 92.7%, but exhibited confusion between specifi cclasses. Hyperparameter tuning and the mixture of experts approach improved the model’s performance. Our proposed model achieved a testing accuracy of 96.25%, precision of 96.28%, and recall of 96.25%, with a 4% increase in testing accuracy, precision, and recall compared to the initial model.
Fair out-of-Distribution detection for addressing skin tone representation in dermatology.
(Université M’hamed Bougara de Boumerdes : Institut de Genie Electrique et Electronique, 2024) Benmalek, ,Assala; Cherifi, Dalila (Supervisor)
Addressing representation issues in dermatological settings is crucial due to variations in how skin conditions manifest across skin tones, thereby providing competitive quality of care across different segments of the population. Although bias and fairness assessment in skin lesion classification has been a nactive research area, there issubstantially less exploration of the implications of skin tone representations and Out-of-Distribution (OOD) detectors’ performance. Current OOD methods detect samples from different hardware devices, clinical settings, or unknown disease samples. However, the absence of robustness analysis across skin tones questions whether these methods are fair detectors. As most skin datasets are reported to suffer from bias in skin tone distribution, this could lead to higher false positive rates in a particular skin tone. This research presents a framework to evaluate OOD detectors across different skin tones and scenarios. We review and compare state-of-the-art OOD detectors across two categories of skin tones, FST I-IV (lighter tones) and FST V-VI (brown and darker tones), over samples collected from dermatoscopic and clinical protocols. We conducted a Gray-Level Co-Occurrence Matrix (GLCM) texture analysis on ”Fitzpatrick17k dataset” samples from two main skin tone categories FST I-IV and FST V-VI, and compared statistical parameters across skin tone categories and nine skin conditions. This analysis indicates that FST V-VI textures are more heterogeneous and varied, while FST I-IV textures are more uniform and consistent. Our OOD detection experiments yield that in poorly performing OOD models, the representation gap measured between skin types is wider (from ? 10% to 30%) up for samples from darker skin tones. Compared to better performing models, skin type performance only differs for? 2%. Furthermore, this work shows that understanding OOD methods’ performance beyond average metrics is critical to developing more fair approaches. We used the AIF360 tool to assess fairness in our OOD detectors and evaluated their performance with group fairness metrics. Our observations show that models with similar overall performance can have significant differences in representation gaps, with group fairness metrics correlating negatively with the representation gap. This indicates that increasing the representation of FST V-VI leads to improved group fairness resulting in fairer OOD detectors.
Exploration of spiking neurones leakages and network recurrences for spike-based-temporal pattern recognition.
(2022) Bouanane, Mohamed Sadek; Cherifi, Dalila (Supervisor)
Brain-inspired computing is being explored to imitate the astonishing capabilities of biological brains to perform robust and efficient computations. To map these capa- bilities into hardware, a growing number of neuromorphic computers are being built to emulate biological neural networks. These developments created a need to ad- dress the lack in understanding of different neuronal behaviours that can enable us to find the right level of abstraction from biology and get the best performance in accurate, efficient and fast inference. Aiming at addressing this problem, we give a detailed overview of the concerned spiking neuron models and surrogate gradient methods, which are used to study the impact of synaptic and membrane leakages in feed-forward, as well as recurrent network topologies on learning visual and auditory information. We also investigate whether or not heterogeneity at the neuronal level plays a functional learning role. We found out that leakages are important when we have both temporal information and a recurrently connected topology. We also found that heterogeneity slightly improves performance on temporal information. The re- sults we obtained will provide more insight on developing efficient neuromorphic hardware.
Improvement of emploitic's search engine results using an ontological approach
(2021) Benbouzid, Souhaib; Bensalah, Abdelkrim; Cherifi, Dalila (Supervisor)
Launched in 2006, Emploitic.com is the leading recruitment website in Algeria. Its goal is to facil-itate job searches for candidates and allow recruiters to ?nd the best pro?les with ease. Emploitic allows users to publish their professional information (training, work history, career summary, so-cial links, etc.) to be used by recruiters when identifying new candidates or to obtain additional information about them. The company achieved its vision by means of a probabilistic information retrieval model developed using the Xapian toolkit. However, as the number of users in the website grew the limits of the old probabilistic model have been reached due to the emergence of a large number of users and their accompanied job posts and resumes. A poor relevance of results retrieved has become noticeable when handling such a sizeable dataset on account of the imitations opposed by the probabilistic model, which are no longer acceptable, which raises the issue to look for new solutions. One prime candidate solution is extending the system with a new service through the use of existing ontology for human resources generated by Emploitic company. Furthermore, the addition of an evaluator software system permitted an assessment of the performance for both old and improved versions of the search engine, showing a clear improvement in the relevance of retrieved documents, in addition, validating the newly developed system as an acceptable solution to this binary classi?cation problem.
DMRI Anatomical Constrained Tractography using proposed tissue maps
(2021) Saoudi, Mazigh; Abdelli, Mohand Saïd; Cherifi, Dalila (Supervisor)
Diffusion Magnetic Resonance Imaging (DMRI), is a technique used to map the brain's inside anatomy in vivo. The images gotten from DMRI can then be passed to reconstruction techniques that estimate the diffusion direction of the fibres at each voxel. Information that can be used to get a 3D model of the brain fibres using tractography algorithms. One inherent limitation of tractography is the determination of the accurate streamline termination point. To address this issue, researches in the field proposed the Anatomical Constrained Tractography (ACT), which is a technique that uses prior knowledge of brain anatomy to restrain the generated streamlines to be biologically coherent. To achieve this, partial volume estimations (PVE) of the different brain's tissues are needed, which are generally segmented from T1 images having a good contrast, especially between the White Matter (WM) and Grey Matter (GM) areas. The contribution of this project is to propose an alternative way to generate the needed PVE’s without using a T1 image. By using diffusion tensor (DT) measures such as Fractional Anisotropy (FA) and Mean Diffusivity (MD), we were able to extract the different brain tissue masks. Our results were promising, they gave masks that are close to the PVE maps provided with the test dataset.
Fiber Tracking Post-processing for abnormal brain tissues analysis using DW-MRI
(Université M’Hamed BOUGARA de Boumerdes : Institut de génie electrique et electronique (IGEE), 2020) Boudjema, Boussaad; Houfaf, Abdessalem; Cherifi, Dalila (Supervisor)
Tractography represents fiber tracts of the cerebral white matter and their connections in the brain based on Di ﬀ usion Weighted Magnetic Resonance Imaging (DW-MRI), which can reveal abnormalities in the white matter especially in ﬁbers’ structure. While this technique is very useful in getting a 3d model of the brain’s neural circuits, it still has some important limitations as it produces a considerable amount of false fibers connections that give rise to false information especially when dealing with abnormal brain tissues. The main objective of our project is to develop an algorithm that aims to improve the accuracy of the tractography results using fiber to bundle coherence measures by cleaning the tractography streamlines and getting rid of spurious fibers and thus obtaining more accurate 3d representation of the brain which will also improve tumors visualization.
A Proposed algorithm for diffusion magnetic resonance imaging brain tractography
(2016) Boudjada, Messaoud; Morsli, Abdelatif; Cherifi, Dalila (Supervisor)
Most of previously published diffusion magnetic resonance imaging reconstruction methods are linked to their own track integration method, some of them use Euler integration others try to improve the integration using the second order and the fourth-order Rang Kutta algorithms. In this work, we have formulated a general, deterministic tractography algorithm (CIERTE), which is a combination of Improved Euler and Range-Kutta fourth-order algorithm, which combine the speed and accuracy at the same time. We have used also trilinear interpolation, which works with voxel level information about fiber orientations including multiple crossings, and employs a range of stopping criteria as those described in EuDX [1] algorithm and FACT [2]. The purpose of this algorithm is to make tracking accurate and fast at the same time and make it general for all deterministic tracking methods.
Face Detection and Recognition in Real Time
(2018) Kheloufi, Anis; Ngheche, Bilal; Cherifi, Dalila (Supervisor)
Face recognition Technology has received significant attention in the past several years in research and application, the existence of several algorithms makes a difficult choice for implementation. The goal of this project is to implement a real time face recognition application, in the first part we implement a system using a global based algorithm (PCA) and we have tested and discussed performance of the application at each level ( detecting, tracking, recognizing). In the second part we tried to make an enhancement of the first algorithm by combining it with two different classifiers, (K-NN and Naïve Bayes), and tested the performance on three different static datasets and discussed results of the

Computer

Browse

Filters

Settings

Sort By

Results per page

Search Results