Adam Cichocki
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Adam Cichocki has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Jaroslaw Tarnawski
Gdansk University of Technology
2 shared publications - Krystian Buszman
Faculty of Navigation and Naval Weapons, Polish Naval Academy, Gdynia, Poland
2 shared publications - Tomasz Adam Rutkowski
Faculty of Electrical and Control Engineering, Gdansk University of Technology, Gdansk, Poland
2 shared publications Miroslaw Woloszyn
Faculty of Electrical and Control Engineering, Gdaqsk University of Technology, Gdaqsk, Poland
1 shared publication- Miroslaw Woloszyn
Faculty of Electrical and Control Engineering, Gdansk University of Technology, Gdansk, Poland
1 shared publication - Stanislaw Osowski
Warsaw University of Technology
1 shared publication
Latest Publications
- Open Access
- 27 January 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 16906-16930
Magnetic Signature Description of Ellipsoid-Shape Vessel Using 3D Multi-Dipole Model Fitted on Cardinal Directions
The article presents a continuation of the research on the 3D multi-dipole model applied to the reproduction of magnetic signatures of ferromagnetic objects. The model structure has been modified to improve its flexibility - model parameters determined by optimization can now be located in the cuboid contour representing the object’s hull. To stiffen the model, the training dataset was expanded to data collected from all four cardinal directions. The robustness of the modified multi-dipole model was verified with various noise levels applied to the synthetic data. A comprehensive numerical verification of the proposed methodology was performed using only data not involved in determining the modified multi-dipole model parameters: the data from intercardinal directions and from different depth were used for cross-validation. An analysis of the influence of initial conditions on the optimization process was carried out. In addition to the gradient optimization method, an evolutionary strategy was also used. Regularization was carried out to search for effective model parameterization. New verification methods were also applied based on the balance of magnetic moments and on the average width of the fit error interval. The results of the performed experiments have shown high robustness of the modified multi-dipole model, even in the face of high noise in the input data. The most significant advantage of the model is its predictive ability, enabling determination of magnetic signatures in any directions and depths with high accuracy.
- 27 January 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 16906-16930
Open AccessMagnetic Signature Description of Ellipsoid-Shape Vessel Using 3D Multi-Dipole Model Fitted on Cardinal Directions
The article presents a continuation of the research on the 3D multi-dipole model applied to the reproduction of magnetic signatures of ferromagnetic objects. The model structure has been modified to improve its flexibility - model parameters determined by optimization can now be located in the cuboid contour representing the object’s hull. To stiffen the model, the training dataset was expanded to data collected from all four cardinal directions. The robustness of the modified multi-dipole model was verified with various noise levels applied to the synthetic data. A comprehensive numerical verification of the proposed methodology was performed using only data not involved in determining the modified multi-dipole model parameters: the data from intercardinal directions and from different depth were used for cross-validation. An analysis of the influence of initial conditions on the optimization process was carried out. In addition to the gradient optimization method, an evolutionary strategy was also used. Regularization was carried out to search for effective model parameterization. New verification methods were also applied based on the balance of magnetic moments and on the average width of the fit error interval. The results of the performed experiments have shown high robustness of the modified multi-dipole model, even in the face of high noise in the input data. The most significant advantage of the model is its predictive ability, enabling determination of magnetic signatures in any directions and depths with high accuracy.
- Open Access
- 16 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 190448-190462
Improving the Quality of Magnetic Signature Reproduction by Increasing Flexibility of Multi-Dipole Model Structure and Enriching Measurement Information
The paper presents the construction of a multi-dipole model that allows reproducing magnetic signatures of ferromagnetic objects. The virtual object used in the paper is an ellipsoid, which is the source of synthetic data. To make the situation more realistic, noise is added to the synthetic data. Two significant improvements compared to previous work are presented. Three-axial magnetometers are introduced instead of uniaxial magnetometers. However, a more important change is the modification of the model structure that allows placing dipoles on the entire plane, e.g. object’s deck. The multi-dipole model consists of an a priori assumed number of permanent and induced single-dipole models. Each single dipole is described by three magnetic moments and, depending on the applied approach, one or two dipole position parameters. The non-linear least-squares optimization method is used to determine model parameters. To assess the quality of magnetic signature reproduction, qualitative and quantitative forms are used. The final quality assessment is based on differences between the reference fields and the fields determined from the multi-dipole model. The applied modifications bring significant improvement, however, only their combined application allows to restore magnetic signatures with good quality for directions other than for which the data were available.
- 16 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 190448-190462
Open AccessImproving the Quality of Magnetic Signature Reproduction by Increasing Flexibility of Multi-Dipole Model Structure and Enriching Measurement Information
The paper presents the construction of a multi-dipole model that allows reproducing magnetic signatures of ferromagnetic objects. The virtual object used in the paper is an ellipsoid, which is the source of synthetic data. To make the situation more realistic, noise is added to the synthetic data. Two significant improvements compared to previous work are presented. Three-axial magnetometers are introduced instead of uniaxial magnetometers. However, a more important change is the modification of the model structure that allows placing dipoles on the entire plane, e.g. object’s deck. The multi-dipole model consists of an a priori assumed number of permanent and induced single-dipole models. Each single dipole is described by three magnetic moments and, depending on the applied approach, one or two dipole position parameters. The non-linear least-squares optimization method is used to determine model parameters. To assess the quality of magnetic signature reproduction, qualitative and quantitative forms are used. The final quality assessment is based on differences between the reference fields and the fields determined from the multi-dipole model. The applied modifications bring significant improvement, however, only their combined application allows to restore magnetic signatures with good quality for directions other than for which the data were available.
- Open Access
- 31 December 2018
- Published by Polish Academy of Sciences Chancellery in Bulletin Of The Polish Academy Of Sciences-Technical Sciences
- Vol. 66 (6) , 757-759
Deep Learning: Theory and Practice
Recent breakthroughs in the fields of artificial intelligence (AI) and machine learning (ML) have been largely triggered by the emergence of the wide class of the deep neural network (DNN) technology, especially of convolutional neural networks (CNN). DNNs have become a vehicle for a large number of potential applications and commercial ventures in computer vision, early diagnosis of some diseases, drug discovery, biomedical informatics, prediction, natural language processing, recommender systems, robotics, gaming and artificial creativity, to mention but a few. The renaissance of deep neural networks has both created an active frontier of research in machine learning and provided many advantages in a variety of applications, to the extent that the performance of DNNs in multi-class classification and verification tasks can be comparable or even better than what is achievable by humans. Deep learning methods continue to dominate the field of machine learning, and are now important in many research areas, especially in artificial intelligence. Thanks to the increasing computational power of computers and the development of new architectures of neural networks, research in this area is greatly accelerated. Deep neural networks, such as convolutional neural networks (CNN) or recurrent long short-term memory (LSTM) networks, have found wide applications in different fields of computer vision and pattern recognition. Examples of such applications include recognition and classification of objects existing in images, image restoration, real-time multi-person pose estimation, computer games, translation, voice generation, music composition, transferring styles from famous paintings, etc. Deep learning was found highly useful in bioengineering, in which we handle the very difficult problems of medical image and signal analysis. This Special Section of the Bulletin of the Polish Academy of Sciences on Technical Sciences is devoted to theoretical aspects of deep machine learning as well as practical applications in some areas of signal and image processing, particularly in bioengineering.
- 31 December 2018
- Published by Polish Academy of Sciences Chancellery in Bulletin Of The Polish Academy Of Sciences-Technical Sciences
- Vol. 66 (6) , 757-759
Open AccessDeep Learning: Theory and Practice
Recent breakthroughs in the fields of artificial intelligence (AI) and machine learning (ML) have been largely triggered by the emergence of the wide class of the deep neural network (DNN) technology, especially of convolutional neural networks (CNN). DNNs have become a vehicle for a large number of potential applications and commercial ventures in computer vision, early diagnosis of some diseases, drug discovery, biomedical informatics, prediction, natural language processing, recommender systems, robotics, gaming and artificial creativity, to mention but a few. The renaissance of deep neural networks has both created an active frontier of research in machine learning and provided many advantages in a variety of applications, to the extent that the performance of DNNs in multi-class classification and verification tasks can be comparable or even better than what is achievable by humans. Deep learning methods continue to dominate the field of machine learning, and are now important in many research areas, especially in artificial intelligence. Thanks to the increasing computational power of computers and the development of new architectures of neural networks, research in this area is greatly accelerated. Deep neural networks, such as convolutional neural networks (CNN) or recurrent long short-term memory (LSTM) networks, have found wide applications in different fields of computer vision and pattern recognition. Examples of such applications include recognition and classification of objects existing in images, image restoration, real-time multi-person pose estimation, computer games, translation, voice generation, music composition, transferring styles from famous paintings, etc. Deep learning was found highly useful in bioengineering, in which we handle the very difficult problems of medical image and signal analysis. This Special Section of the Bulletin of the Polish Academy of Sciences on Technical Sciences is devoted to theoretical aspects of deep machine learning as well as practical applications in some areas of signal and image processing, particularly in bioengineering.