Researcher: Saul, Can Jozef
Name Variants
Saul, Can Jozef
Email Address
Birth Date
3 results
Search Results
Now showing 1 - 3 of 3
Publication Metadata only Abalone life phase classification with deep learning(IEEE, 2018) Ozsarfati, Eran; Yılmaz, Alper; N/A; N/A; Şahin, Egemen; Saul, Can Jozef; Researcher; Researcher; N/A; N/A; N/A; N/AIn this paper, we present algorithmic and architectural comparison of deep learning models for predicting abalone age range. While abalone age can be determined through very detailed steps in a laboratory, we present an efficient method for determining its age through machine learning models. We present a precise and an efficient method for converting data to a computable version through binary encoding and normalization. We experiment with various topological variances in neural network architectures, convolutional approach to the task at hand and recently succeeding residual neural network architecture for finding the optimal prediction accuracy and efficiency. Although the conventional machine learning methods showed success in this field, our deep learning model tests yield an accuracy of 79.09% accuracy, surpassing the conventional machine learning algorithms as we incorporated methods for preventing over-fitting and methods for normalizing the output throughout the network.Publication Metadata only Carbon nanotube coordinate prediction with deep learning(Institute of Electrical and Electronics Engineers Inc., 2019) N/A; N/A; Saul, Can Jozef; Şahin, Egemen; Researcher; Researcher; N/A; N/A; N/A; N/AThe development of carbon nanotube technology has provided a great advantage in applications of many fields including nanotechnology and materials science due to the exquisite mechanical, chemical, thermal and electrical properties of carbon nanotubes. However, due to their size, the scale at which the physical phenomena of carbon nanotubes are apparent is too small to do physical experiments, there is a need for certain computational methods like molecular dynamics simulations. In this present study, we propose a deep learning methodology, alongside a custom data preprocessing method, for precisely determining carbon nanotubes’ coordinates. We experimented with various topologies of neural networks and acquired a top result of 81.34%. Our findings and computation method surpasses the previous work on this field, in terms accuracy and computational time.Publication Metadata only Carbon nanotube coordinate prediction with deep learning(Ieee, 2019) Kamış, Ayşe Zeynep; N/A; N/A; Saul, Can Jozef; Şahin, Egemen; Researcher; Researcher; N/A; N/A; N/A; N/AThe development of carbon nanotube technology has provided a great advantage in applications of many fields including nanotechnology and materials science due to the exquisite mechanical, chemical, thermal and electrical properties of carbon nanotubes. However, due to their size, the scale at which the physical phenomena of carbon nanotubes are apparent is too small to do physical experiments, there is a need for certain computational methods like molecular dynamics simulations. In this present study, we propose a deep learning methodology, alongside a custom data preprocessing method, for precisely determining carbon nanotubes' coordinates. We experimented with various topologies of neural networks and acquired a top result of 81.34%. Our findings and computation method surpasses the previous work on this field, in terms accuracy and computational time.