An Experiment on the Use of Genetic Algorithms for Topology Selection in Deep Learning

Mattioli, Fernando; Caetano, Daniel; Cardoso, Alexandre; Naves, Eduardo Lázaro Martins; Lamounier, Edgard

doi:10.1155/2019/3217542

Cited by 13 publications

(8 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…By implementing a proximity sensor, all virtual objects within range are rendered, as in the Poke-mon GO game [44]. Otherwise, the use of deep learning techniques can be investigated in order to reduce rendering failure [45]. Studies will be conducted, to improve the control signals and video stream latency.…”

Section: Discussionmentioning

confidence: 99%

The Augmented Reality Telerehabilitation System for Powered Wheelchair User’s Training

Caetano

Valentini²,

Mattioli

et al. 2020

JCIS

Self Cite

View full text Add to dashboard Cite

Many people worldwide have been experimenting a decrease in their mobility as a result of aging, accidents and degenerative diseases. In many cases, a Powered Wheelchair (PW) is an alternative help. Currently in Brazil, patients can receive a PW from the Unified Health System, following prescription criteria. However, they do not have an appropriate previous training for driving the PW. Consequently, users might suffer accidents since a customized training protocol is not available. Nevertheless, due to financial and/or health limitations, many users are unable to attend a rehabilitation center. To overcome these limitations, we developed an Augmented Reality Telerehabilitation System Architecture based on the Power Mobility Road Test (PMRT) for supporting PW user's training. In this system, the therapists can remotely customize and evaluate training tasks and the user can perform the training in safer conditions. The video stream and data transfer between each environment were made possible through UDP (User Datagram Protocol). To evaluate and present the system architecture potential, a preliminary test was conducted with 3 spinal cord injury participants. They performed 3 basic training protocols defined by a therapist. The following metrics were adopted for evaluation: number of control commands; elapsed time; number of collisions; biosignals. Also, a questionary was used to evaluate system features. Preliminary results demonstrated the specific needs of individuals using a PW thanks to adopted metrics (qualitative and emotional). Results have shown the need for a training system with customizable protocols to fulfill these needs. User's evaluation demonstrates that the combination of AR techniques with PMRT adaptations, increases user's well-being after training sessions. Furthermore, a training experience helps users to overcome their displacement problems, as well as for appointing challenges before large scale use. This system architecture allows further studies on telerehabilitation of PW users training.

show abstract

Section: Discussionmentioning

confidence: 99%

The Augmented Reality Telerehabilitation System for Powered Wheelchair User’s Training

Caetano

Valentini²,

Mattioli

et al. 2020

JCIS

Self Cite

View full text Add to dashboard Cite

show abstract

“…The above drawbacks render the method of uninformed search inferior to that of informed search, as shown by [18], [35], [36]. The informed search uses and learns the knowledge obtained from previous results.…”

Section: A Optimization Of Hyperparameters and Architecturementioning

confidence: 99%

“…Most of the recent studies on deep neuroevolution are applied to image classification [10], [11], [13]- [18] and only a few to text classification [12], [19], [20]. Text data faces different challenges as compared to image data, as it has a large number of words, which leads to sparsity and high computational cost [43], [44].…”

Section: B Deep Neuroevolutionmentioning

confidence: 99%

An Automatic Convolutional Neural Network Optimization Using a Diversity-Guided Genetic Algorithm

Fatyanosa

Aritsugi

2021

IEEE Access

View full text Add to dashboard Cite

Hyperparameters and architecture greatly influence the performance of convolutional neural networks (CNNs); therefore, their optimization is important to obtain the desired results. One of the state-ofthe-art methods to achieve this is the use of neuroevolution that utilizes a genetic algorithm (GA) to optimize a CNN. However, the GA is often trapped into a local optimum resulting in premature convergence. In this study, we propose an approach called the "diversity-guided genetic algorithm-convolutional neural network (DGGA-CNN)" that uses adaptive parameter control and random injection to facilitate the search process by exploration and exploitation while preserving the population diversity. The alternation between exploration and exploitation is guided by using an average pairwise Hamming distance. Moreover, the DGGA fully handles the architecture of the CNN by using a novel finite state machine (FSM) combined with three novel mutation mechanisms that are specifically created for architecture chromosomes. Tests conducted on suggestion mining and twitter airline datasets reveal that the DGGA-CNN performs well with valid architectures and a comparison with other methods demonstrates its capability and efficiency.

show abstract

“…Nigam et al successfully conducted a study using a hybrid GA and DNN for molecule design 24 . Mattioli et al explained the use of GAs in DNN topology selection 25 . Kononenko 26 studied machine learning methods for medical diagnosis and diagnostics.…”

Section: Introductionmentioning

confidence: 99%

Classification of some chemical drugs by genetic algorithm and deep neural network hybrid method

Karakaplan

Avcu

2021

Concurrency and Computation

View full text Add to dashboard Cite

Summary Deep neural networks (DNN) and genetic algorithm (GA) are gaining importance quickly with many successful applications in the field of science and technology. They are indispensable tool for the numerical solution of difficult problems. It is possible to optimize DNNs using the GA and this combination can be used to classify data. In this article, some drugs are classified by Monte Carlo sampling with combination of GA and DNN due to stochastic nature of the domain, exponential number of variables and small number of chemical species. In addition to the values obtained from the databases of selected drugs, molecular dynamic and ab initio molecular mechanical calculation results were also used. The aim of this study is to generalize the molecular classification with the data obtained from chemical databases as well as molecular docking results by using the combination of deep learning and GA and its usability in drug design. The selected drugs are some agonist and antagonist drugs that bind to dopamine receptors, which are widely studied and well known in the literature. To train the DNN, input datasets were chosen by the GA framework written in pure Python named PyEvolve. Classification of drugs has been analyzed with the focus on orbital energies and docking results. It is possible to use this algorithm in many in silico calculations such as affinity and separation processes. The reliability of the algorithm was tested with the results given in the literature and the expected values were estimated at 93.8%.

show abstract

An Experiment on the Use of Genetic Algorithms for Topology Selection in Deep Learning

Cited by 13 publications

References 11 publications

The Augmented Reality Telerehabilitation System for Powered Wheelchair User’s Training

The Augmented Reality Telerehabilitation System for Powered Wheelchair User’s Training

An Automatic Convolutional Neural Network Optimization Using a Diversity-Guided Genetic Algorithm

Classification of some chemical drugs by genetic algorithm and deep neural network hybrid method

Contact Info

Product

Resources

About