Research

Past

My masters level research focused on topics in machine learning including deep learning, neural network chaining, and hybrid symbolic-connectionist models. I have applied these techniques successfully in industry in process control and time-series prediction, and academically in natural language processing and navigation and collision avoidance.

My master's thesis, Sequential decision making using a two stage hybrid connectionist model explored the concept of using multiple neural network models simultaneously to both speed learning and acquire knowledge in real time (as applied to navigation).

Post graduation I continued research in the area of hybrid symbolic-connectionist models, combining symbolic models (expert systems) to constrain and guide connectionist models (neural networks). In the course of this work I was principal investigator and technical author of the patent Method and apparatus for optimizing operation of a power generating plant using artificial intelligence techniques, which directly led to a $3M US DOE research grant. Following the award, I was lead architect and intelligent systems engineer, designing and building a process control system for coal-fired power plant optimization based on the same patent. This system was successfully installed in multiple power plants, reducing pollution and fuel costs while holding or increasing power output.

Current and Future

Adviser: Professor Vangelis Metsis, PhD

My research interests are in supervised machine learning as applied to complex time-series data and temporal classification (e.g. biometrics, physiometrics, process control, temporal event prediction. etc.). I also have strong interests in the use of machine learning on the Web, high performance computing, and software engineering.

Publications

• Master's Thesis: Sequential decision making using a two stage hybrid connectionist model
• Method and apparatus for optimizing operation of a power generating plant using artificial intelligence techniques
• A Preliminary Experimental Analysis on RateMyProfessors. Byron Gao and Alexander Katrompas. 2020 IEEE International Conference on Big Data (Big Data).
• Rate My Professors: A Study Of Bias and Inaccuracies In Anonymous Self-Reporting. Alexander Katrompas and Vangelis Metsis. International Conference on Computing and Data Science 2021.
• Enhancing LSTM Models with Self-Attention and Stateful Training. Alexander Katrompas, Vangelis Metsis, Intelligent Systems Conference (IntelliSys) 2021.
• Recurrence and Self-Attention vs the Transformer for Time-Series Classification: A Comparative Study. Alexander Katrompas, Theodoros Ntakouris, Vangelis Metsis. 2022 International Conference on Artificial Intelligence in Medicine, Published in by Springer Nature, 2022.
• Temporal Attention for Improved Time Series Classification and Interpretability. Alexander Katrompas, Vangelis Metsis. International Conference on Artificial Neural Networks 2023, Published in by Springer Nature, 2023.
• Many-to-Many Prediction for Effective Modeling of Frequent Label Transitions in Time Series. Alexander Katrompas, Vangelis Metsis. PETRA'24, ACM ISBN 979-8-4007-1760-4/24/06.
• Recurrence And Temporal Attention Synergy For Optimal Time-Series Modelling And Interpretability. PhD Dissertation, Texas State University. Alexander Katrompas, Vangelis Metsis. Passed October 10th, 2023. Available December 2024.

Just For Fun

if(its==!false) its="funny";