Scientific Seminar of the Department: Schedule 2026

• 28-04-2026. A guest lecturer, Desislava Vasileva, will present a research overview on Adaptive Educational Software and Game-Based Learning in hall 278 at IMI-Sofia and online from 2:00 p.m. She will talk on adaptive educational software and game-based learning, covering three interconnected research lines. The first concerns adaptive hypermedia and e-learning systems, including formal learner modeling and a platform for adaptive content delivery. The second focuses on game-based learning and player modeling — playing style classification from behavioral data, physiological affect detection via galvanic skin response, and a platform for automated generation of adaptive 3D educational games. The third extends this work to serious games for cultural heritage and STEM education, with ongoing research on multi-modal player modeling combining eye-tracking, galvanic skin response and behavioral signals for personalized secondary school games.
• 31-03-2026. Opening of the seminar for 2026 will start at 14.00 (UTC + 02), hall 278 of IMI-BAS, Sofia and online via ZOOM
  Assist. PhD student Ventsislav Polimenov will give a talk, entitled “ANNSIA – Adaptive Neural Network for Satellite Image Analysis“.
  The report presents an innovative approach to leaf area index estimation using a multi-sensor deep neural network that combines Sentinel-2 and Landsat 8/9 data. The study addresses two primary challenges, namely the absence of in-situ data for training and the substantial differences in spatial and spectral characteristics between the sensors. The model employs a Multi-Sensor U-Net with Conditional Batch Normalisation, enabling unified processing of heterogeneous sensor inputs and the learning of sensor-agnostic vegetation representations. Training is conducted through a VI-ensemble approach that generates pseudo-LAI labels without requiring field measurements. Data balancing strategies and spatial block validation ensure reliable assessment of model accuracy and generalisation capacity.
  The results demonstrate improved model performance when leveraging information from both satellites, as well as successful transfer to new geographic regions and time periods. This constitutes a contribution to the still underexplored field of multi-sensor remote sensing and establishes a foundation for future extensions to other biophysical variables and the integration of additional satellite data sources.