John B. Theocharis was born in Trikala, Greece in 1956. He received the diploma and the PhD degree in electrical Engineering from the Dept. of Electrical Engineering, Faculty of Engineering, Aristotle University of Thessaloniki, Greece, in 1979 and 1985, respectively. Since 1980, he has been with the Department of Electrical and Computer Engineering (ECE) at the Aristotle University of Thessaloniki (AUTh), Greece, where he is currently a Full Professor on Dynamic Systems modeling.

His current research interests include: Computational intelligence, Fuzzy Systems, Neural Networks, Evolutionary algorithms, Machine learning, Deep learning, Modeling/Classification, Pattern recognition, Medical Imaging. He has authored over 140 publications in scientific journals and conference proceedings.

His duties include teaching undergraduate courses in Electrical Circuit theory, Automatic Control systems I, Computational Intelligence and Active filter design. He also offers graduate courses in Computational intelligence – Bio-inspired systems: Fuzzy systems, Evolutionary Computation- Bio-inspired algorithms, and Machine learning techniques.

He has served as director of the Division of Electrical and Computer Engineering of the ECE Department and the Control and robotics Lab. He is currently director of the postgraduate program in “Advanced computer and communication systems”. Prof. J. B. Theocharis is a Member of the IEEE and the Technical Chamber of Greece.


Prof. John B. Theocharis
Aristotle University of Thessaloniki,
Department of Electrical and Computer Engineering
Division of Electronics and Computer Engineering
54124 Thessaloniki, Greece.
Tel.: +30 2310 996343
Fax: +30 2310 996447
Email: theochar(at),


Research interests

  • Machine learning: Development of classification/regression models based on SVMs and graph-based data representation. Sparse representation techniques, Dimensionality reduction methods, Transfer learning and Active learning approaches. Methods are applied for classification/segmentation of hyperspectral remotely sensed images, soil properties prediction in soil spectroscopy, and medical imaging: knee MRI image segmentation using multi-atlas approach, osteoarthritis severity prediction from MRI images.
  • Deep learning: Convolutional Neural Networks (CNNs) for image classification and system’s modeling. Stacked autoencoders (SAEs) for better feature space representation. The techniques are applied in soil spectroscopy, registration/segmentation and osteoarthritis prediction from MRI knee images.
  • Soil spectroscopy: Soil properties analysis from Vis-NIR spectral signatures. Development of highly interpretable Fuzzy rule-based models with feature selection capabilities using efficient evolutionary search algorithms (Differential evolution, DE). Application of graph-based Lap-SVR models combined with Active learning in soil analysis. Methods are investigated in large soil data bases and especially the European LUCAS data library.
  • Fuzzy modeling: Development of hybrid neuro-fuzzy models in the form of IF/THEN rules, combining fuzzy inference principles and the adaptation qualities of Neural Networks. Parameter/structure learning algorithms for the extraction of parsimonious fuzzy models.
  • Recurrent Fuzzy Neural Networks (RFNN): Modeling of non-linear and non-stationary processes via RFNN, with fuzzy partition of the antecedent part and inclusion of dynamics in the consequent part. Investigation of different forms of recurrence, using external feedback paths or dynamic neurons in chained forms.
  • Static/Recurrent Neural Networks: Research on multilayered recurrent neural networks with internal local dynamics, where the synaptic links are implemented via IIR filters. Effective techniques for the gradient computation and on-line learning algorithms (global/decoupled) for optimal adaptation of recurrent network weights.
  • Genetic algorithms and evolutionary strategies: Efficient GA schemes are used for non-linear constraint optimization problems and structure/parameter learning of neuro-fuzzy models.
  • Neuro-Fuzzy Classifiers: Construction of hierarchical multilayered classifiers with self-organizing properties (SONeFMUC), where generic neuro-fuzzy classifiers are interconnected along multiple layers. Novel classifier structures incorporating classifiers combination, decision fusion, data splitting and classification improvement of ambiguous patterns. Efficient structure learning techniques integrating diversity measures in the classifier’s ensembles.
  • Fuzzy Decision Tree Classifiers: New hierarchical tree classifiers, where the node separations are implemented via binary SVMs. Effective structure learning strategies are developed incorporation a class grouping algorithm based fuzzy confusion metrics and individual feature selection per node.
  • Genetic Fuzzy Rule-Based Classification Systems (GRBCS): Evolutionary search algorithms for the extraction of GFRBCSs with compact structures and high interpretability properties, under the iterative rule learning approach. Use of GFRBCSs for land cover classification from multispectral satellite images.
  • Biomedical/biomechanical Engineering: Image classification and segmentation based on computational intelligence techniques for biomedical (tissue characterization) and biomechanical (gait analysis and pathologies detection) applications.
  • Feature extraction/selection: Development of constructive feature selection techniques to identify discriminating features for classification tasks. Novel methods (SVM-FuzzCoC) integrating the fuzzy partition vector of features, feature search based on a complementary criterion and support vector machines.
  • Adaptive control: Development of adaptive neuron-fuzzy controllers for uncertain system’s control. Control laws and adaptation rules for the network weights assuring uniform ultimate boundedness of the tracking.


Application  Fields 

  • Adjustable speed drives of AC machines using neural network state estimators.
  • Short-term load forecasting in the Greek interconnected power system, using neuro-fuzzy models.
  • Long-term wind speed and power prediction in wind farms using recurrent neural networks and recurrent fuzzy neural networks.
  • Genetic Algorithm resolution of the hydrothermal coordination and production scheduling of the Greek Power system.
  • Separation of pathological discontinuous adventitious sounds form vesicular sounds related to pulmonary dysfunctions, using static/ recurrent neural networks.
  • Land-cover, forest species and burned areas classification form multi-spectral and hyperspectral remoted sensed imagery, using neuro-fuzzy classifiers, fuzzy decision tree classifiers and GFRBCSs.
  • Tissue characterization of Intravascular Ultrasound (IVUS) images using feature extraction/selection techniques and support vector machines.
  • Subject recognition system based ground reaction forces of human gaits, combining wavelet packet decomposition, feature selection and support vector machines.
  • Investigation of osteoarthritis severity from ground reaction force measurements, using wavelet subband extraction and fuzzy decision tree classifiers.