Pub Date : 2019-05-01DOI: 10.1109/ICCAIRO47923.2019.00015
A. Tashk, J. Herp, E. Nadimi
Colorectal lesions known as polyps are one of the diagnostic symptoms for colorectal disease. So, their accurate detection and localization based on a computer-aided diagnosis can assist colonists for prescribing more effective treatments. The computer vision and machine learning methods like pattern recognition and deep learning neural networks are the most popular strategies for automatic polyp detection purpose. The proposed approach in this paper is an innovative deep learning neural network. The proposed network has a novel U-Net architecture. The architecture of proposed network includes fully 3D layers which enable the network to be fed with multi or hyperspectral images or even video streams. Moreover, there is a dice prediction output layer. This type of output layer employs probabilistic approaches and benefits from more accurate prediction abilities. The proposed method is applied to international standard optical colonoscopy datasets known as CVC-ClinicDB, CVC-ColonDB and ETIS-Larib. The implementation and evaluation results demonstrate that the proposed U-Net outperforms other competitive methods for automatic polyp detection based on accuracy, precision, recall and F-Score criteria. The proposed method can assist experts and physicians to localize colonial polyps with more accuracy and speed. In addition, the proposed network can be used on live colonoscopy observations due to its high performance and fast operability.
{"title":"Fully Automatic Polyp Detection Based on a Novel U-Net Architecture and Morphological Post-Process","authors":"A. Tashk, J. Herp, E. Nadimi","doi":"10.1109/ICCAIRO47923.2019.00015","DOIUrl":"https://doi.org/10.1109/ICCAIRO47923.2019.00015","url":null,"abstract":"Colorectal lesions known as polyps are one of the diagnostic symptoms for colorectal disease. So, their accurate detection and localization based on a computer-aided diagnosis can assist colonists for prescribing more effective treatments. The computer vision and machine learning methods like pattern recognition and deep learning neural networks are the most popular strategies for automatic polyp detection purpose. The proposed approach in this paper is an innovative deep learning neural network. The proposed network has a novel U-Net architecture. The architecture of proposed network includes fully 3D layers which enable the network to be fed with multi or hyperspectral images or even video streams. Moreover, there is a dice prediction output layer. This type of output layer employs probabilistic approaches and benefits from more accurate prediction abilities. The proposed method is applied to international standard optical colonoscopy datasets known as CVC-ClinicDB, CVC-ColonDB and ETIS-Larib. The implementation and evaluation results demonstrate that the proposed U-Net outperforms other competitive methods for automatic polyp detection based on accuracy, precision, recall and F-Score criteria. The proposed method can assist experts and physicians to localize colonial polyps with more accuracy and speed. In addition, the proposed network can be used on live colonoscopy observations due to its high performance and fast operability.","PeriodicalId":297342,"journal":{"name":"2019 International Conference on Control, Artificial Intelligence, Robotics & Optimization (ICCAIRO)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126312786","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 : 2019-05-01DOI: 10.1109/ICCAIRO47923.2019.00036
Dženan Gušić, Z. Šabanac, Sanela Nesimović
In this paper we complement the most recent results on soundness of inference rules for new vague multivalued dependencies. Motivated by the fact that the inclusive and the augmentation rules are sound, we prove that: complementation, transitivity, replication, coalescence, union, pseudo-transitivity, decomposition, and mixed pseudo-transitivity rules are also sound. Our research relies on definitions of vague functional and vague multivalued dependencies based on appropriately selected similarity measures between vague values, vague sets, and tuples on sets of attributes.
{"title":"On Soundness of Various Inference Rules for Vague Functional Dependencies","authors":"Dženan Gušić, Z. Šabanac, Sanela Nesimović","doi":"10.1109/ICCAIRO47923.2019.00036","DOIUrl":"https://doi.org/10.1109/ICCAIRO47923.2019.00036","url":null,"abstract":"In this paper we complement the most recent results on soundness of inference rules for new vague multivalued dependencies. Motivated by the fact that the inclusive and the augmentation rules are sound, we prove that: complementation, transitivity, replication, coalescence, union, pseudo-transitivity, decomposition, and mixed pseudo-transitivity rules are also sound. Our research relies on definitions of vague functional and vague multivalued dependencies based on appropriately selected similarity measures between vague values, vague sets, and tuples on sets of attributes.","PeriodicalId":297342,"journal":{"name":"2019 International Conference on Control, Artificial Intelligence, Robotics & Optimization (ICCAIRO)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131305679","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 : 2019-05-01DOI: 10.1109/ICCAIRO47923.2019.00043
D. Tarasov, O. Milder
Fine-tuning the reproduction of the initial colorants pure colors gradations is the basis of color reproduction in modern printing systems. Usually, tone reproduction curves are constructed by successively changing the tone of the basic dyes (CMYK). However, this approach does not take into account the effect of changes in the dyes shade when they overlap. As an alternative basis for color correction, we previously suggested using gradation trajectories, which are analogous to gradation curves in the CIE Lab space. We also proposed a discrete approach to computing them, using natural color discretization in digital printing devices. In this article, we propose to use three-dimensional gradation surfaces in the CIE Lab space as a mathematical model of double color overlays (RGB) and as a further development of the idea of gradation trajectories. The calculations use the mathematical apparatus of the differential geometry of spatial curves and surfaces. The color space metric is determined by the value of the CIE dE color difference. To simplify the application of the model, we also propose to carry out calculations in discrete form. In this case, color coordinates are considered as continuous functions of filling a discrete raster cell with two dyes. As gradation trajectories, we consider geodesic lines on the gradation surfaces of the corresponding double overlaps of dyes. For calculations we also used a discrete approach. Experimental verification was carried out using an electrophotographic printer.
{"title":"Discrete Gradation Surfaces Computation in Electrophotography","authors":"D. Tarasov, O. Milder","doi":"10.1109/ICCAIRO47923.2019.00043","DOIUrl":"https://doi.org/10.1109/ICCAIRO47923.2019.00043","url":null,"abstract":"Fine-tuning the reproduction of the initial colorants pure colors gradations is the basis of color reproduction in modern printing systems. Usually, tone reproduction curves are constructed by successively changing the tone of the basic dyes (CMYK). However, this approach does not take into account the effect of changes in the dyes shade when they overlap. As an alternative basis for color correction, we previously suggested using gradation trajectories, which are analogous to gradation curves in the CIE Lab space. We also proposed a discrete approach to computing them, using natural color discretization in digital printing devices. In this article, we propose to use three-dimensional gradation surfaces in the CIE Lab space as a mathematical model of double color overlays (RGB) and as a further development of the idea of gradation trajectories. The calculations use the mathematical apparatus of the differential geometry of spatial curves and surfaces. The color space metric is determined by the value of the CIE dE color difference. To simplify the application of the model, we also propose to carry out calculations in discrete form. In this case, color coordinates are considered as continuous functions of filling a discrete raster cell with two dyes. As gradation trajectories, we consider geodesic lines on the gradation surfaces of the corresponding double overlaps of dyes. For calculations we also used a discrete approach. Experimental verification was carried out using an electrophotographic printer.","PeriodicalId":297342,"journal":{"name":"2019 International Conference on Control, Artificial Intelligence, Robotics & Optimization (ICCAIRO)","volume":"49 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126928465","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 : 2019-05-01DOI: 10.1109/ICCAIRO47923.2019.00022
Lilija Naiwert, K. Spindler
This paper, which grew out of an ongoing project geared towards improvements in control education, provides an example of an autonomous two-dimensional control system which is simple enough to be completely analyzed analytically, but rich enough to exhibit interesting features. This example can be readily used in a course on optimal control to illustrate und visualize relevant control-theoretical concepts. We tried to present the material in a comprehensible and illustrative way which can be easily adapted for classroom use.
{"title":"Explicit Calculation of Reachable Sets to Illustrate Concepts in Optimal Control","authors":"Lilija Naiwert, K. Spindler","doi":"10.1109/ICCAIRO47923.2019.00022","DOIUrl":"https://doi.org/10.1109/ICCAIRO47923.2019.00022","url":null,"abstract":"This paper, which grew out of an ongoing project geared towards improvements in control education, provides an example of an autonomous two-dimensional control system which is simple enough to be completely analyzed analytically, but rich enough to exhibit interesting features. This example can be readily used in a course on optimal control to illustrate und visualize relevant control-theoretical concepts. We tried to present the material in a comprehensible and illustrative way which can be easily adapted for classroom use.","PeriodicalId":297342,"journal":{"name":"2019 International Conference on Control, Artificial Intelligence, Robotics & Optimization (ICCAIRO)","volume":"126 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128584653","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 : 2019-05-01DOI: 10.1109/ICCAIRO47923.2019.00020
I. C. Demetriou, Ioannis N. Perdikas
We present examples of peak estimation to measurements of Raman, Infrared and NMR spectra by the piecewise monotonic data approximation method. The structural differences of these spectra, the complexity of the underlying physical laws and the error included in the measurements make this a good test of the effectiveness of the method. Precisely, if a number of monotonic sections of the data is required, then the optimal turning points and the least sum of squares of residuals are computed in quadratic complexity with respect to the number of data. This is a remarkable result because the problem may require an enormous number of combinations in order to find the optimal turning point positions. Our results exhibit some strengths and indicate certain advantages of the method. Therefore, they may be helpful to the development of new algorithms that are particularly suitable for peak estimation in spectroscopy calculations.
{"title":"The Effectiveness of the Piecewise Monotonic Approximation Method for the Peak Estimation of Noisy Univariate Spectra","authors":"I. C. Demetriou, Ioannis N. Perdikas","doi":"10.1109/ICCAIRO47923.2019.00020","DOIUrl":"https://doi.org/10.1109/ICCAIRO47923.2019.00020","url":null,"abstract":"We present examples of peak estimation to measurements of Raman, Infrared and NMR spectra by the piecewise monotonic data approximation method. The structural differences of these spectra, the complexity of the underlying physical laws and the error included in the measurements make this a good test of the effectiveness of the method. Precisely, if a number of monotonic sections of the data is required, then the optimal turning points and the least sum of squares of residuals are computed in quadratic complexity with respect to the number of data. This is a remarkable result because the problem may require an enormous number of combinations in order to find the optimal turning point positions. Our results exhibit some strengths and indicate certain advantages of the method. Therefore, they may be helpful to the development of new algorithms that are particularly suitable for peak estimation in spectroscopy calculations.","PeriodicalId":297342,"journal":{"name":"2019 International Conference on Control, Artificial Intelligence, Robotics & Optimization (ICCAIRO)","volume":"66 12","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133008998","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 : 2019-05-01DOI: 10.1109/ICCAIRO47923.2019.00032
András Formanek, D. Hadhazi
Convolutional Neural Networks have recently taken over the field of image processing, because they can handle complex non algorithmic problems with state-of-the-art results, based on precision and inference times. However, there are many environments (e.g. cell phones, IoT, embedded systems, etc.) and use-cases (e.g. pedestrian detection in autonomous driving assistant systems), where the hard real-time requirements can only be satisfied by efficient computational resource utilization. The general trend is training larger and more complex networks in order to achieve better accuracies and forcing these networks to be redundant (in order to increase their generalization ability). However, this produces networks that cannot be used in such scenarios. Pruning methods try to solve this problem by reducing the size of the trained neural networks. These methods eliminate the redundant computations after the training, which usually cause high drop in the accuracy. In this paper, we propose new regularization techniques, which induce the sparsity of the parameters during the training and in this way, the network can be efficiently pruned. From this viewpoint, we analyse and compare the effect of minimizing different norms of the weights (L1, L0) one by one and for groups of them (for kernels and channels). L1 regularization can be optimized by Gradient Descent, but this is not true for L0. The paper proposes a combination of Proximal Gradient Descent optimization and RMSProp method to solve the resulting optimization problem. Our results demonstrate that the proposed L0 minimization-based regularization methods outperform the L1 based ones, both in terms of sparsity of the resulting weight-matrices and the accuracy of the pruned network. Additionally, we demonstrate that the accuracy of deep neural networks can also be increased using the proposed sparsifying regularizations.
{"title":"Compressing Convolutional Neural Networks by L0 Regularization","authors":"András Formanek, D. Hadhazi","doi":"10.1109/ICCAIRO47923.2019.00032","DOIUrl":"https://doi.org/10.1109/ICCAIRO47923.2019.00032","url":null,"abstract":"Convolutional Neural Networks have recently taken over the field of image processing, because they can handle complex non algorithmic problems with state-of-the-art results, based on precision and inference times. However, there are many environments (e.g. cell phones, IoT, embedded systems, etc.) and use-cases (e.g. pedestrian detection in autonomous driving assistant systems), where the hard real-time requirements can only be satisfied by efficient computational resource utilization. The general trend is training larger and more complex networks in order to achieve better accuracies and forcing these networks to be redundant (in order to increase their generalization ability). However, this produces networks that cannot be used in such scenarios. Pruning methods try to solve this problem by reducing the size of the trained neural networks. These methods eliminate the redundant computations after the training, which usually cause high drop in the accuracy. In this paper, we propose new regularization techniques, which induce the sparsity of the parameters during the training and in this way, the network can be efficiently pruned. From this viewpoint, we analyse and compare the effect of minimizing different norms of the weights (L1, L0) one by one and for groups of them (for kernels and channels). L1 regularization can be optimized by Gradient Descent, but this is not true for L0. The paper proposes a combination of Proximal Gradient Descent optimization and RMSProp method to solve the resulting optimization problem. Our results demonstrate that the proposed L0 minimization-based regularization methods outperform the L1 based ones, both in terms of sparsity of the resulting weight-matrices and the accuracy of the pruned network. Additionally, we demonstrate that the accuracy of deep neural networks can also be increased using the proposed sparsifying regularizations.","PeriodicalId":297342,"journal":{"name":"2019 International Conference on Control, Artificial Intelligence, Robotics & Optimization (ICCAIRO)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130831024","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 : 2019-05-01DOI: 10.1109/ICCAIRO47923.2019.00021
R. Matousek, T. Hulka, Ladislav Dobrovsky, J. Kůdela
Symbolic Regression (SR) is a well-studied method in Genetic Programming (GP) for discovering free-form mathematical models from observed data, which includes not only the model parameters but also its innate structure. Another level of the regression problem is the design of an appropriate fitness function, by which are individual solutions judged. This paper proposes a new fitness function design for symbolic regression problems called a Sum epsilon-Tube Error (STE). The function of this criterion can be visualized as a tube with a small radius that stretches along the entire domain of the approximated function. The middle of the tube is defined by points that match approximated valued (in the so-called control points). The evaluation function then compares, whether each approximated point does or does not belong to the area of the tube and counts the number of points outside of the epsilon-Tube. The proposed method is compared with the standard sum square error in several test cases, where the advantages and disadvantages of the design are discussed. The obtained results show great promise for the further development of the STE design and implementation.
{"title":"Sum Epsilon-Tube Error Fitness Function Design for GP Symbolic Regression: Preliminary Study","authors":"R. Matousek, T. Hulka, Ladislav Dobrovsky, J. Kůdela","doi":"10.1109/ICCAIRO47923.2019.00021","DOIUrl":"https://doi.org/10.1109/ICCAIRO47923.2019.00021","url":null,"abstract":"Symbolic Regression (SR) is a well-studied method in Genetic Programming (GP) for discovering free-form mathematical models from observed data, which includes not only the model parameters but also its innate structure. Another level of the regression problem is the design of an appropriate fitness function, by which are individual solutions judged. This paper proposes a new fitness function design for symbolic regression problems called a Sum epsilon-Tube Error (STE). The function of this criterion can be visualized as a tube with a small radius that stretches along the entire domain of the approximated function. The middle of the tube is defined by points that match approximated valued (in the so-called control points). The evaluation function then compares, whether each approximated point does or does not belong to the area of the tube and counts the number of points outside of the epsilon-Tube. The proposed method is compared with the standard sum square error in several test cases, where the advantages and disadvantages of the design are discussed. The obtained results show great promise for the further development of the STE design and implementation.","PeriodicalId":297342,"journal":{"name":"2019 International Conference on Control, Artificial Intelligence, Robotics & Optimization (ICCAIRO)","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116626575","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：