Pub Date : 2021-01-21DOI: 10.1109/SAMI50585.2021.9378684
M. Gruosso, N. Capece, U. Erra, Flavio Biancospino
In recent years, deep reinforcement learning has increasingly contributed to the development of robotic applications and boosted research in robotics. Deep learning and model-free, off-policy, value-based reinforcement learning algorithms enabled agents to successfully learn complex robotic skills through trial and error process and visual inputs. The aim of this paper concerns the training of a robot in a simulation environment by designing a Deep Q-Network (DQN) that elaborates images acquired by an RGB vision sensor inside a 3D simulated environment and outputs a value for each action the robotic arm can execute given the current state. In particular, the robot has to push a ball into a soccer net without any knowledge of the environment and its own location. In addition, our further goal was to perform agent validation during training and assess its generalization level. Despite the many advances in reinforcement learning, it is still a challenge. Therefore, we devised a validation strategy similar to the method applied in supervised learning and tested the agent both on known and unknown experiences, achieving interesting and promising results.
{"title":"A Validation Approach for Deep Reinforcement Learning of a Robotic Arm in a 3D Simulated Environment","authors":"M. Gruosso, N. Capece, U. Erra, Flavio Biancospino","doi":"10.1109/SAMI50585.2021.9378684","DOIUrl":"https://doi.org/10.1109/SAMI50585.2021.9378684","url":null,"abstract":"In recent years, deep reinforcement learning has increasingly contributed to the development of robotic applications and boosted research in robotics. Deep learning and model-free, off-policy, value-based reinforcement learning algorithms enabled agents to successfully learn complex robotic skills through trial and error process and visual inputs. The aim of this paper concerns the training of a robot in a simulation environment by designing a Deep Q-Network (DQN) that elaborates images acquired by an RGB vision sensor inside a 3D simulated environment and outputs a value for each action the robotic arm can execute given the current state. In particular, the robot has to push a ball into a soccer net without any knowledge of the environment and its own location. In addition, our further goal was to perform agent validation during training and assess its generalization level. Despite the many advances in reinforcement learning, it is still a challenge. Therefore, we devised a validation strategy similar to the method applied in supervised learning and tested the agent both on known and unknown experiences, achieving interesting and promising results.","PeriodicalId":402414,"journal":{"name":"2021 IEEE 19th World Symposium on Applied Machine Intelligence and Informatics (SAMI)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133018507","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-21DOI: 10.1109/SAMI50585.2021.9378618
Szabolcs Csaholczi, L. Kovács, L. Szilágyi
The segmentation of brain tumor and the separation of its parts like the enhancing core or edema represents a highly important problem, since a fine solution offers precise diagnosis and better opportunities in radiotherapy planning or follow-up studies after interventions. Brain tumor segmentation is also a highly challenging task, due to the wide variety of lesion appearances, the possible presence of noise effects, and the differences in MRI scanner sensitivity. This paper is a preliminary study of a random forest (RF) based solution for the tumor part segmentation problem using multi-spectral MRI data. The proposed method is trained and tested using the 220 high-grade glioma records of the BraTS 2015 train data set. These records are preprocessed to eliminate noise effects and to generate 100 additional features to the four observed ones. The output of the RF classifier is fed directly to statistical evaluation, in order to investigate the direct contribution of the RF to the accurate segmentation. The overall Dice scores exceeding 82% for the whole tumor, 80% for the enhancing core, 74% for the tumor core, and 72% for the edema, make the random forest classifier a good candidate to be successful as the core of a multistage brain tumor part segmentation procedure.
{"title":"Automatic Segmentation of Brain Tumor Parts from MRI Data Using a Random Forest Classifier","authors":"Szabolcs Csaholczi, L. Kovács, L. Szilágyi","doi":"10.1109/SAMI50585.2021.9378618","DOIUrl":"https://doi.org/10.1109/SAMI50585.2021.9378618","url":null,"abstract":"The segmentation of brain tumor and the separation of its parts like the enhancing core or edema represents a highly important problem, since a fine solution offers precise diagnosis and better opportunities in radiotherapy planning or follow-up studies after interventions. Brain tumor segmentation is also a highly challenging task, due to the wide variety of lesion appearances, the possible presence of noise effects, and the differences in MRI scanner sensitivity. This paper is a preliminary study of a random forest (RF) based solution for the tumor part segmentation problem using multi-spectral MRI data. The proposed method is trained and tested using the 220 high-grade glioma records of the BraTS 2015 train data set. These records are preprocessed to eliminate noise effects and to generate 100 additional features to the four observed ones. The output of the RF classifier is fed directly to statistical evaluation, in order to investigate the direct contribution of the RF to the accurate segmentation. The overall Dice scores exceeding 82% for the whole tumor, 80% for the enhancing core, 74% for the tumor core, and 72% for the edema, make the random forest classifier a good candidate to be successful as the core of a multistage brain tumor part segmentation procedure.","PeriodicalId":402414,"journal":{"name":"2021 IEEE 19th World Symposium on Applied Machine Intelligence and Informatics (SAMI)","volume":"2 11","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133043070","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-21DOI: 10.1109/SAMI50585.2021.9378647
Dávid Magyar, S. Szénási
Since programming languages exist, parsing the code - primarily for compilation - is an ever-present necessity. As of today, there is a deep theory of this topic, describing approaches and useful constructs, categories and capabilities of parsers, which this paper not intends to dive into deeply, but to describe and use the official technical terms where possible. The presented approach of parsing utilizes regular expressions and forms a PEG (Parsing Expression Grammar), which is more expressive than simply regular expressions [1]. This paper aims to present an approach specially for parsing complex input recursively using PEG approach. An easy to configure and understand interpreter based on regular expressions over characters, tokens and schemas is outlined.
{"title":"Parsing via Regular Expressions","authors":"Dávid Magyar, S. Szénási","doi":"10.1109/SAMI50585.2021.9378647","DOIUrl":"https://doi.org/10.1109/SAMI50585.2021.9378647","url":null,"abstract":"Since programming languages exist, parsing the code - primarily for compilation - is an ever-present necessity. As of today, there is a deep theory of this topic, describing approaches and useful constructs, categories and capabilities of parsers, which this paper not intends to dive into deeply, but to describe and use the official technical terms where possible. The presented approach of parsing utilizes regular expressions and forms a PEG (Parsing Expression Grammar), which is more expressive than simply regular expressions [1]. This paper aims to present an approach specially for parsing complex input recursively using PEG approach. An easy to configure and understand interpreter based on regular expressions over characters, tokens and schemas is outlined.","PeriodicalId":402414,"journal":{"name":"2021 IEEE 19th World Symposium on Applied Machine Intelligence and Informatics (SAMI)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115716663","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-21DOI: 10.1109/SAMI50585.2021.9378646
Atakan Kaya, Kubilay Atas, I. Myderrizi
The number of COVID-19 patients around the globe is increasing day by day. Statistics show that even after almost 10 months from outbreak, number of the total patients has not reached to its peak value yet. Easy spreading of the virus among people causes high number of patients at the same time. Accelerating the reduction in spread is of vital importance. In order to achieve this reduction, early diagnosis of the disease and the number of tests and scans to be performed frequently becomes important. In this paper, a comprehensive model examination is made to overcome COVID-19 diagnosing problem. Using CT images, data augmentation technique is applied first in the pre-processing section and then pre-trained deep CNN networks perform the classification. The model is tested using various networks and high accuracy results of 96.5% and 97.9% are obtained for VGG-16 and EfficientNetB3 networks, respectively.
{"title":"Implementation of CNN based COVID-19 classification model from CT images","authors":"Atakan Kaya, Kubilay Atas, I. Myderrizi","doi":"10.1109/SAMI50585.2021.9378646","DOIUrl":"https://doi.org/10.1109/SAMI50585.2021.9378646","url":null,"abstract":"The number of COVID-19 patients around the globe is increasing day by day. Statistics show that even after almost 10 months from outbreak, number of the total patients has not reached to its peak value yet. Easy spreading of the virus among people causes high number of patients at the same time. Accelerating the reduction in spread is of vital importance. In order to achieve this reduction, early diagnosis of the disease and the number of tests and scans to be performed frequently becomes important. In this paper, a comprehensive model examination is made to overcome COVID-19 diagnosing problem. Using CT images, data augmentation technique is applied first in the pre-processing section and then pre-trained deep CNN networks perform the classification. The model is tested using various networks and high accuracy results of 96.5% and 97.9% are obtained for VGG-16 and EfficientNetB3 networks, respectively.","PeriodicalId":402414,"journal":{"name":"2021 IEEE 19th World Symposium on Applied Machine Intelligence and Informatics (SAMI)","volume":"117 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116185164","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-21DOI: 10.1109/sami50585.2021.9378685
R. Hudák, M. Schnitzer, J. Živčák
Nowadays, additive manufacturing otherwise known as three-dimensional (3D) printing is fully implemented into the production of hard tissue replacements. Department of Biomedical Engineering and Measurement together with Biomedical Engineering company designed and produced more than 300 implants made of Ti64 ELI titanium alloy using additive technologies, which were subsequently implanted by surgeons worldwide. 3D printing of PEEK, bioceramic and magnesium alloys implants is recently tested to offer alternative materials to titanium for cranioplasties or biodegradable impalnts. 3D bioprinting is being applied to regenerative medicine to address the need for tissues and organs suitable for transplantation. Compared with non-biological printing, 3D bioprinting involves additional complexities, such as the choice of materials, cell types, growth and differentiation factors, and technical challenges related to the sensitivities of living cells and the construction of tissues. The 3D bioplotter was used to prepare tubular structures made of PLA + PHB polymer for substitutes of human urethra. Tubular structures were tested from geometrical point of view to assure required precision, repeatability and possibility to print porous structures for application of epithelial and muscle cells and their growth. Several studies on PEEK spinal implants manufactured by 3D printing were realized, where mechanical testing, simulations and testing of biocompatibility were implemented. Presented research covers selected case studies of patient specific implants made by additive manufacturing and research in medical 3D bioprinting for tissue engineering.
{"title":"Additive Manufacturing in Medicine and Tissue Engineering: Plenary Talk","authors":"R. Hudák, M. Schnitzer, J. Živčák","doi":"10.1109/sami50585.2021.9378685","DOIUrl":"https://doi.org/10.1109/sami50585.2021.9378685","url":null,"abstract":"Nowadays, additive manufacturing otherwise known as three-dimensional (3D) printing is fully implemented into the production of hard tissue replacements. Department of Biomedical Engineering and Measurement together with Biomedical Engineering company designed and produced more than 300 implants made of Ti64 ELI titanium alloy using additive technologies, which were subsequently implanted by surgeons worldwide. 3D printing of PEEK, bioceramic and magnesium alloys implants is recently tested to offer alternative materials to titanium for cranioplasties or biodegradable impalnts. 3D bioprinting is being applied to regenerative medicine to address the need for tissues and organs suitable for transplantation. Compared with non-biological printing, 3D bioprinting involves additional complexities, such as the choice of materials, cell types, growth and differentiation factors, and technical challenges related to the sensitivities of living cells and the construction of tissues. The 3D bioplotter was used to prepare tubular structures made of PLA + PHB polymer for substitutes of human urethra. Tubular structures were tested from geometrical point of view to assure required precision, repeatability and possibility to print porous structures for application of epithelial and muscle cells and their growth. Several studies on PEEK spinal implants manufactured by 3D printing were realized, where mechanical testing, simulations and testing of biocompatibility were implemented. Presented research covers selected case studies of patient specific implants made by additive manufacturing and research in medical 3D bioprinting for tissue engineering.","PeriodicalId":402414,"journal":{"name":"2021 IEEE 19th World Symposium on Applied Machine Intelligence and Informatics (SAMI)","volume":"67 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122628265","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-21DOI: 10.1109/SAMI50585.2021.9378656
L. Vavrek, Matej Hires, D. Kumar, P. Drotár
This paper describes the investigation of the use of the deep neural networks (DNN) for the detection of pathological speech. The state-of-the-art VGG16 convolutional neural network based transfer learning was the basis of this work and different approaches were trialed. We tested the different architectures using the Saarbrucken Voice database (SVD). To overcome limitations due to language and education, the SVD was limited to /a/, /i/ and /u/ vowel subsets with sustained natural pitch. The scope of this study was only diseases that classify as organic dysphonia. We utilized multiple simple networks trained separately on different vowel subsets and combined them as a single model ensemble. It was found that model ensemble achieved an accuracy on pathological speech detection of 82 %. Thus, our results show that pre-trained convolutional neural networks can be used for transfer learning when input is the spectrogram representation of the voice signal. This is significant because it overcomes the need for very large data size that is required to train DNN, and is suitable for computerized analysis of the speech without limitation of the language skills of the patients.
{"title":"Deep convolutional neural network for detection of pathological speech","authors":"L. Vavrek, Matej Hires, D. Kumar, P. Drotár","doi":"10.1109/SAMI50585.2021.9378656","DOIUrl":"https://doi.org/10.1109/SAMI50585.2021.9378656","url":null,"abstract":"This paper describes the investigation of the use of the deep neural networks (DNN) for the detection of pathological speech. The state-of-the-art VGG16 convolutional neural network based transfer learning was the basis of this work and different approaches were trialed. We tested the different architectures using the Saarbrucken Voice database (SVD). To overcome limitations due to language and education, the SVD was limited to /a/, /i/ and /u/ vowel subsets with sustained natural pitch. The scope of this study was only diseases that classify as organic dysphonia. We utilized multiple simple networks trained separately on different vowel subsets and combined them as a single model ensemble. It was found that model ensemble achieved an accuracy on pathological speech detection of 82 %. Thus, our results show that pre-trained convolutional neural networks can be used for transfer learning when input is the spectrogram representation of the voice signal. This is significant because it overcomes the need for very large data size that is required to train DNN, and is suitable for computerized analysis of the speech without limitation of the language skills of the patients.","PeriodicalId":402414,"journal":{"name":"2021 IEEE 19th World Symposium on Applied Machine Intelligence and Informatics (SAMI)","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129891067","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-21DOI: 10.1109/SAMI50585.2021.9378633
K. Beneda
Although propulsion systems in commercial aviation rely on high bypass ratio turbofan engines, there is still a niche in which turbojet engines can be utilized. Despite some promising experiments that offer the on-wing measurement of the most important parameter, the thrust of the engine is still not available during flight. Turbofan Power Ratio, which is a compound thermodynamic value of various pressures and temperatures across the engine, is proportional to the thrust output of the turbofan, and the same relationship was proven by the author earlier regarding turbojet engines with fixed geometry exhaust nozzle. This paper has the main objective to gather data that can reveal how variable geometry affects the relationship between Turbofan Power Ratio and thrust output of the turbojet. This has been performed by carrying out measurements on a real turbojet engine test bed. The results show that the correlation is not suitable directly to determine thrust levels as it is influenced by nozzle position. Therefore, the author has developed a novel thrust parameter that is derived from TPR and can provide additional diagnostic capabilities. The outcome of this research can gain additional importance in the future as several engine manufacturers are about to introduce variable geometry nozzles, and the results presented in this paper may pave the way for these succeeding developments.
{"title":"Investigation of Novel Thrust Parameters to Variable Geometry Turbojet Engines","authors":"K. Beneda","doi":"10.1109/SAMI50585.2021.9378633","DOIUrl":"https://doi.org/10.1109/SAMI50585.2021.9378633","url":null,"abstract":"Although propulsion systems in commercial aviation rely on high bypass ratio turbofan engines, there is still a niche in which turbojet engines can be utilized. Despite some promising experiments that offer the on-wing measurement of the most important parameter, the thrust of the engine is still not available during flight. Turbofan Power Ratio, which is a compound thermodynamic value of various pressures and temperatures across the engine, is proportional to the thrust output of the turbofan, and the same relationship was proven by the author earlier regarding turbojet engines with fixed geometry exhaust nozzle. This paper has the main objective to gather data that can reveal how variable geometry affects the relationship between Turbofan Power Ratio and thrust output of the turbojet. This has been performed by carrying out measurements on a real turbojet engine test bed. The results show that the correlation is not suitable directly to determine thrust levels as it is influenced by nozzle position. Therefore, the author has developed a novel thrust parameter that is derived from TPR and can provide additional diagnostic capabilities. The outcome of this research can gain additional importance in the future as several engine manufacturers are about to introduce variable geometry nozzles, and the results presented in this paper may pave the way for these succeeding developments.","PeriodicalId":402414,"journal":{"name":"2021 IEEE 19th World Symposium on Applied Machine Intelligence and Informatics (SAMI)","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133193562","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：