Marek Bundzel, Ph.D.
Born 1974 in Slovakia.
Ph.D.: : (2005) Structural and Parametrical Adaptation of Artificial Neural Networks Using Principles of Support Vector Machines (pdf)
Ph.D. research advisor: Assoc. Prof. Dr. Peter Sincak
Technical University of Kosice, Slovakia.
MSc.: (1997) Application of Neural Networks in Water Management
Technical University of Kosice, Slovakia
Languages: English - fluently, German - advanced, Russian - advanced
General: Methods of computational intelligence, Artificial Neural Networks (ANN, supervised and unsupervised learning), Support Vector Machines (SVM, I was giving lectures on Support Vector Machines at the workshop of the Fourth International Conference on Hydroinformatics, Iowa City, USA), Evolutionary algorithms (EA). Development of methods for structural and parametrical learning of Artificial Neural Networks (gradient and evolutionary optimization of ANN’s). Using ANN and SVM for prediction (urban runoff prediction during storm events, error correction of a physical model of a water system, financial predictions etc.) and for pattern recognition (land use identification in satellite images etc.). Development of a method for ensembling classifiers using unsupervised learning (for distributed computing). Application of computational geometry algorithms in development of SVM alternative. Application of evolutionary optimization for development of ANN controllers for simulated robots (evolutionary robotics). Experimental work with Lego NXT platform in the domain of evolutionary robotics.
Fellowship at Waseda University, Tokyo (2 years): The memory-prediction framework (a theory of brain function that was created by Jeff Hawkins and described in his 2004 book On Intelligence): Development of a model of the memory-prediction framework designed specifically for the purposes of object recognition in mobile robot. The resulting model uses unsupervised learning to discover objects in the robot’s world. The human operator can give names to the objects found. The model identifies spatial and temporal patterns in the visual data sensed by a moving robot. Robotics: Design (including 3D CAD), manufacture (including CNC machining) and assembly of a mobile observation robot (including installation of electrical and electronic components and programming). The robot features include 13 sonars, 4 cameras, 2 encoders, ports for additional sensors, 3 microcontrollers and a mobile Dell XPS computer for master control, WIFI communication and interfacing with the operator (audio-visual). Design and manufacture of a universal self-aligning coupling mechanism for reconfigurable robots (prototype, 65mm diameter).
Present research (since Dec. 2010): Further development of the model of memory-prediction framework, development of a generalized version of the model for the purposes of financial forecasting. Experimental work with humanoid robot NAO (Aldebaran Robotics), installation of a robot soccer control software.
Software and Downloads
Memory Prediction Framework Model for Unsupervised Learning of a Mobile Robot (2010)
In 2004, Jeff Hawkins presented a memory-prediction theory of brain function, and later used it to create the Hierarchical Temporal Memory model. Several of the concepts described in the theory are applied here in a computer vision system for a mobile robot application. The aim was to produce a system enabling a mobile robot to explore its envi-ronment and recognize different types of objects without human supervision. The operator has means to assign names to the identified objects of interest. The system presented here works with time ordered sequences of images. It utilizes a tree structure of connected computational nodes similar to Hierarchical Temporal Memory and memorizes frequent sequences of events. This is a beta version, a proper documented version is on the way please stand by. More information on the system.
Evolutionary Optimization of a Novel Tonearm (2006)
This is a set of Matlab scripts running evolutionary optimization of the parameters of a high end tonearm of a novel design. The purpose of the design is to maintain the reading needle in the optimal angle to the grooves on the record (tangent).
SVM Matlab Toolbox (2000)
A bit outdated but still useful for regression and pattern recognition on smaller datasets. Documentation included.
3D Construction Model of a Mobile Observation Robot (2010)
This SketchUp model describes almost all mechanical components (including camera casings and wheels) of a differential wheeled observation robot. I have designed and build the robot as a testing platform for autonomous learning based on visual data and specifically for testing of the above mentioned memory-prediction system. I have machined the components using a 3 axis CNC mill at Waseda University, Tokyo.
The robot is propelled by two RDO-37KE50G9A motors (geared, with encoder). The motors are driven by Sabertooth dual 12A motor driver. The encoders' pulses are counted by two Microchip PIC32 microcontrollers (hardware pulse counting feature was needed to accurately count the pulses from the encoders, 9000 per one wheel rotation). The Microchip PIC32 microcontrollers pass the odometry information to the Arduino Mega microcontroller board via a serial link. The robot is equipped with 13 Maxbotix MaxSonar ultrasonic rangefinders (five MB1210 XL MaxSonar EZ1 in the front upper row, three in the rear, five MB1230 XL MaxSonar EZ3 in the front bottom row). Arduino Mega is programmed to control the motors, to collect the sensory information (odometry, rangefinders) and to communicate with the main computer - Dell XPS M1330 notebook. There is a Sanwa 7 port USB 2.0 hub installed in the robot. Dell XPS, Arduino Mega and two Logitech Webcam Pro 9000 connect to the Sanwa hub. New aluminium casings were made for the webcams' boards enabling installation of the cameras in a stereoscopic setup primarily in the front of the robot. The Logitech webcams were chosen for their wide field of view and high image quality. When mounted in their primary position the optical axes of the lenses are parallel to each other (50mm apart) and to the floor (100mm above). Each camera covers 61-degree horizontal and 48-degree vertical field of view. The vertical field of view touches the floor 230mm in front of the robot. The horizontal fields of view of the cameras start to overlap 43mm in front of the robot. The robot is powered by a 12V battery. The Dell XPS battery powers only the notebook itself. (www links as of June 2011).
The image data were recorded when moving the mount in a 1.7m x 2.1m arena.
3D Construction Model of a Coupling for Reconfigurable Modular Robots (2010)
This SketchUp model describes the mechanical components of coupling for Reconfigurable Modular Robots. The coupling was designed and the prototype was build as the pilot work for a research project proposal. The idea was to make a coupling enabling the modules to be hot plugged any to any to form many different structures. Opposite to the male-female design, the proposed couplings have uniform design enabling any module to be plugged into any free coupling. The couplings provide sturdy mechanical and electrical connection. Connecting the coupling pair is possible in several rotated positions. The coupling pair engages by pressing the couplings together till a mechanical lock clicks into place. The design of the coupling enables limited self alignment during the rendezvous thus facilitating the (re)configuration process. The couplings are equipped with an actuator to release the lock. The components of the prototype have been machined using a 3 axis CNC mill at Waseda University, Tokyo.
As a teaching assistant (labs): Artificial Neural Networks, Evolutionary Computation, Computational Intelligence, Artificial Intelligence. Programming in C, Applied Programming in Windows (Visual Studio, C#).
Complete courses (lectures and labs): Biocybernetics and Evolutionary Robotics, Artificial Life.
Course development: Biocybernetics and Evolutionary Robotics (Evolutionary robotics - development, Biocybernetics – update), Artificial Life (Update)
Supervision of Master Projects
- Marek Skokan: Pseudo-distance Based Artificial Neural Network Training.
- Michal Majza: Application of Intelligent Technologies in Weather Forecast.
- Martin Brziak: Using Particle Swarm Optimization for Structural and Parametrical Adaptation of Feedforward ANN.
- Martin Sramko: Application of Echo State Networks in Prediction.
- Mikulas Darabos: Dynamic Adaptation of Topology of Neural Network Ensemble.
- Peter Hrobak: Application of Cellular Automata in Pattern Recognition.
- Imrich Knut: Application of ANN in Recognition of Music Styles.
Attila Miliczky: AdaBoost - How Strong Should Weak Classifiers Be?
- Gabriel Toth: Evolution of Movement Gaits for Snake Robot Using Neuro-controller.
- Tomas Sladik: Evolution of Walking Gait for Bipedal Walking Robot.
- Marian Onder: Implementation and Analysis of the FLY Algorithm for Processing of Stereoscopic Images
- Dusan Holodak: Evolution of a CPG based Neuro-controller for Bipedal Walking Robot
- Tomas Kocis: Evolution of 3D Movement Gaits for Snake Robot Using Neuro-controller.
- Martin Wolf: Navigation of Mobile Robot Using Convolutional Networks and Stereovision
- Using methods of computational intelligence in Smart Grids
- Using methods of computational intelligence in Smart Ticketing
1997 – Technical University of Budapest, Hungary (1 month)
1998 – University of Pavia, Italy (1 month)
1999,2000 – Danish Hydraulic Institute, Denmark (7 months)
2008-2010 - SHALAB, Waseda University, Tokyo (Postdoc, 2 years).
Bundzel, M., Hashimoto S., (2010). Object Identification in Dynamic Images Based on the Memory-Prediction Theory of Brain Function, Journal of Intelligent Learning Systems and Applications, vol. 2, pp. 212-220, ISSN 2150-8402.
Bundzel, M., and Sinčák, P.,(2008). Ensembling Classifiers Using Unsupervised Learning. In Artificial Intelligence and Soft Computing ICAISC 2008, 9th International Conference, series: Lecture Notes in Computer Science, vol. 5097, pp. 513-521, ISBN 978-3-540-69572-1.
Bundzel, M., and Kasanicky, T. (2008). Growing Ensemble of Classifiers. In 6th international symposium on applied machine intelligence and informatics SAMI 2008, (pp. 183-187).
Skokan, M., Bundzel, M., and Sincak, P. (2008). Pseudo-distance Based Artificial Neural Network Training. In 6th international symposium on applied machine intelligence and informatics SAMI 2008, (pp. 59-62).
Bundzel, M., and Sincak, P. (2006). Combining Gradient and Evolutionary Approaches to the Artificial Neural Networks Training According to Principles of Support Vector Machines. IEEE Int. Joint Conf. on Neural Networks (IJCNN2006), Vancouver, Canada, ISBN: 0-7803-9490-9.
Bundzel, M., Kasanicky, T., and Frankovic, B. (2006). Building Support Vector Machine Alternative Using Algorithms of Computational Geometry. IEEE International Symposium on Neural Networks 2006, Chengdu, China, series Lecture Notes in Computer Science, Advances in Neural Networks - ISNN 2006, ISBN978-3-540-34439-1, ISSN0302-9743, Springer Berlin / Heidelberg, (pp. 955-961).
Bundzel, M., and Kasanicky, T. (2006). Using Algorithms of Computational Geometry for Pattern Recognition and Comparison to Support Vector Machine. In: Znalosti 2006 : 5. rocnik konference, VSB-TU Ostrava, ISBN 80-248-1001-8, (pp. 60-70).
Bundzel, M., Sinčák, P., and Kopco, N. (2000), Using Support Vector Machines for Classification of Remotely Sensed Images, in Proceedings of the European Symposium on Computational Intelligence, Kosice, Slovakia
Babovic, V., Keijzer, M., and Bundzel, M. (2000). From global to local modeling: A case study in error correction of deterministic models, in Proceedings of the Fourth International Conference on Hydroinformatics, Iowa City, USA
Bundzel, M., Sinčák, P., Sokáč, M., Sztruhár, D., Marsalek, J. (1998). Urban Runoff Prediction by Neural Networks. In Proceedings of the Third International Conference on Hydroinformatics, Babovic & Larsen (eds.), Balkema, Rotterdam, ISBN 90 5410 983 1
Bundzel, M., Sincak, P., Sokac, M., Sztruhar, D., Marsalek, J. (1999). Hydroinformatic Tools in City Hydrology (in Slovak: Vyuzitie prostriedkov hydroinformatiky v mestskej hydrologii). (UVTIP Press, ISBN 80-85330-64-4).
Babovic, V., Keijzer, M., Bundzel, M. (2000). Error Correction in Venice Lagoon using Neural Networks, Support Vector Machines and Local Linear Models, , D2K Technical Report, Danish Hydraulic Institute, Horsholm, Denmark.
Bundzel., M. (1999). D2K Support Vector Machine MATLAB toolbox, D2K Technical Report, Danish Hydraulic Institute, Horsholm, Denmark.
2000: lectures on Support Vector Machines in tutorials of the Fourth International Conference on Hydroinformatics, Iowa City, USA.
Bundzel, M. (2008). Parallelization of Pattern Recognition Using Ensembling with Unsupervised Learning. Twelfth International Conference on Cognitive and Neural Systems, Boston University, 677 Beacon Street, Boston, Massachusetts 02215 USA.
Bundzel, M. (2005). Structural and parametrical adaptation of artificial neural networks using principles of support vector machines. Ninth International Conference on Cognitive and Neural Systems, Boston University, 677 Beacon Street, Boston, Massachusetts 02215 USA.
Invited lectures for Japanese high school students under JSPS Science Dialoque:
2010, Yamanashi Prefectural Tsuru High School, Japan, “On Computional Intelligence”. 2009, "Science Camp" at Takasaki High School in Gunma, “On Computional Intelligence”.