LVQ neural network optimized implementation on FPGA devices with multiple-wordlength operations for real-time systems

Blaiech, Ahmed Ghazi; Khalifa, Khaled Ben; Boubaker, Mohamed; Bedoui, Mohamed Hédi

doi:10.1007/s00521-016-2465-7

Cited by 13 publications

(4 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…So to switch from SOM to LVQ, it is simply a matter of modifying the SOMPE' architecture (Figure 4) by removing the neighbourhood unit and adding an additional specific input to the expert's label that is necessary for supervised learning. This architecture may well be used in an electroencephalogram (EEG) classification application already published in [25] but adopting a different architectural approach.…”

Section: Discussionmentioning

confidence: 99%

A Novel Hardware Systolic Architecture of a Self-Organizing Map Neural Network

Khalifa

Blaiech

Bedoui

2019

Computational Intelligence and Neuroscience

Self Cite

View full text Add to dashboard Cite

In this article, we propose to design a new modular architecture for a self-organizing map (SOM) neural network. The proposed approach, called systolic-SOM (SSOM), is based on the use of a generic model inspired by a systolic movement. This model is formed by two levels of nested parallelism of neurons and connections. Thus, this solution provides a distributed set of independent computations between the processing units called neuroprocessors (NPs) which define the SSOM architecture. The NP modules have an innovative architecture compared to those proposed in the literature. Indeed, each NP performs three different tasks without requiring additional external modules. To validate our approach, we evaluate the performance of several SOM network architectures after their integration on an FPGA support. This architecture has achieved a performance almost twice as fast as that obtained in the recent literature.

show abstract

Section: Discussionmentioning

confidence: 99%

A Novel Hardware Systolic Architecture of a Self-Organizing Map Neural Network

Khalifa

Blaiech

Bedoui

2019

Computational Intelligence and Neuroscience

Self Cite

View full text Add to dashboard Cite

show abstract

“…EEG signal processing steps with HPO of DL models for vigilance state classification vasive acquisition process, with electrodes placed along the scalp. To prepare the dataset, we use the same two subjects (S1, S2) as those collected in the experimentation of the previous work of our team [14] [5]. The EEG data are directly recorded from 28 active electrodes from the scalp at the Department of Functional Explorations of the Nervous System at Sahloul University Hospital, Tunisia.…”

Section: Eeg Signal Acquisition and Pre-processingmentioning

confidence: 99%

“…In the first step of preprocessing, we split the signal into time periods of four seconds (recommended by an expert), in order to reduce the computation complexity. Then, we filter this signal to eliminate artifacts using a high-pass filter to remove low frequencies less than 0.1 Hz, and a low-pass filter to filter out frequencies above 21Hz, in order to focus on frequencies most related to the state of alertness.Experts agree that this range is one of the most relevant ranges for vigilance.The next step of pre-processing is the spectral analysis of the signal which was proposed in [14] [5]:…”

Section: Eeg Signal Acquisition and Pre-processingmentioning

confidence: 99%

Hyperparameter Optimization of Deep Learning Models for EEG-Based Vigilance Detection

Khessiba

Blaiech

Manzanera

et al. 2022

Advances in Computational Collective Intelligence

Self Cite

View full text Add to dashboard Cite

ElectroEncephaloGraphy (EEG) signals have a nonlinear and complex nature and require the design of sophisticated methods for their analysis. Thus, Deep Learning (DL) models, which have enabled the automatic extraction of complex data features at high levels of abstraction, play a growing role in the field of medical science to help diagnose various diseases, and have been successfully used to predict the vigilance states of individuals. However,the performance of these models is highly sensitive to the choice of the hyper-parameters that define the structure of the network and the learning process. When targeting an application, tuning the hyper-parameters of deep neural networks is a tedious and time-consuming process.This explains the necessity of automating the calibration of these hyper-parameters. In this paper, we perform hyperparameters optimization using two popular methods: Tree Parzen Estimator (TPE) and Bayesian optimisation (BO) to predict vigilance states of individuals based on their EEG signal. The performance of the methods is evaluated on the vigilance states classification. Compared with empirical optimization,the accuracy is improved from 0.84 to 0.93 with TPE and from 0.84 to 0.97 with Bayesian optimization using the 1D-UNet-LSTM deep learning model. Obtained results show that the combination of the 1D-UNet encoder and LSTM offers an excellent compromise between the performance and network size (thus training duration), which allows a more efficient hyper-parameter optimization.

show abstract

“…In medical images, we need to deal with millions or billion pixels per picture to recognize or diagnose particular diseases. Thus, a huge number of computational processes are needed at the same time and the run time should be taken into consideration also [12][13][14][15].…”

Section: Introductionmentioning

confidence: 99%

Enhancing the Speed of the Learning Vector Quantization (LVQ) Algorithm by Adding Partial Distance Computation

AbuAlghanam

Adwan

Shariah

et al. 2022

Cybernetics and Information Technologies

View full text Add to dashboard Cite

Learning Vector Quantization (LVQ) is one of the most widely used classification approaches. LVQ faces a problem as when the size of data grows large it becomes slower. In this paper, a modified version of LVQ, which is called PDLVQ is proposed to accelerate the traditional version. The proposed scheme aims to avoid unnecessary computations by applying an efficient Partial Distance (PD) computation strategy. Three different benchmark datasets are used in the experiments. The comparisons have been done between LVQ and PDLVQ in terms of runtime and in result, it turns out that PDLVQ shows better efficiency than LVQ. PDLVQ has achieved up to 37% efficiency in runtime compared to LVQ when the dimensions have increased. Also, the enhanced algorithm (PDLVQ) shows clear enhancement to decrease runtime when the size of dimensions, the number of clusters, or the size of data becomes increased compared with the traditional one which is LVQ.

show abstract

LVQ neural network optimized implementation on FPGA devices with multiple-wordlength operations for real-time systems

Cited by 13 publications

References 21 publications

A Novel Hardware Systolic Architecture of a Self-Organizing Map Neural Network

A Novel Hardware Systolic Architecture of a Self-Organizing Map Neural Network

Hyperparameter Optimization of Deep Learning Models for EEG-Based Vigilance Detection

Enhancing the Speed of the Learning Vector Quantization (LVQ) Algorithm by Adding Partial Distance Computation

Contact Info

Product

Resources

About