Melih Talha Bektaş introduced his artificial intelligence research at an international conference

Melih Talha Bektaş, a second-year student in the Electrical-Electronics Engineering Department of Üsküdar University’s Faculty of Engineering and Natural Sciences, introduced his Cross-Gate approach developed in the field of artificial intelligence at UbiComp/ISWC 2026 held in Shanghai, China. The research, which focuses on how features learned from different sensors can be transferred between two artificial intelligence models, yielded positive results in initial validation experiments conducted in the field of human activity recognition. 

Melih Talha Bektaş, a second-year student in the Electrical-Electronics Engineering Department of Üsküdar University’s Faculty of Engineering and Natural Sciences, introduced his research in the field of artificial intelligence at the ACM International Joint Conference on Pervasive and Ubiquitous Computing and International Symposium on Wearable Computing (UbiComp/ISWC 2026) held in Shanghai, China.

Learnable Connections Between Two Artificial Intelligence Models…

Bektaş's research titled “What Does a Trained Gate Actually Transfer Between Two Models?” examines how two neural networks trained with different sensor data can share the features they have learned. The Cross-Gate approach, central to the study, aims to establish connections between the learned feature representations of two different models trained with infrared (IR) and depth (Depth) sensor data, rather than merely combining their final predictions.

Developed with a Three-Stage Training Approach

The mechanism, based on learnable linear contributions, was developed with a three-stage training approach. Within the scope of the research, six different video models were evaluated on the task of recognizing 40 different human activities. In initial validation experiments, when end-to-end training was applied, an improvement was observed in four out of six model configurations compared to the simple prediction averaging method. In the study, which achieved an average gain of approximately 2.2 percentage points, research aimed at validating the findings with more comprehensive experiments is ongoing.

“I Am Interested in the Changes Occurring Within the Models”

Explaining the main motivation of his research, Bektaş stated that he focuses not only on recognizing human movements more accurately but also on understanding how the internal representations of neural networks trained with different data interact with each other. Describing his approach to artificial intelligence, Bektaş said, “I am generally interested not only in the input and output of an artificial intelligence model but also in the changes occurring within the model during learning.”

Stating that the Cross-Gate study is an applied example of this approach, Bektaş noted that he aims to establish connections between the features learned by different sensors with low additional computational cost and to understand their effects on learning.

Scientific Contact with International Researchers

During the conference, Bektaş had the opportunity to meet academics from various universities and scientists from research teams, and stated that his discussions with researchers from universities in Hong Kong, in particular, were an important experience for him.

“An Important Step in My Research Journey”

Bektaş, reminding that he is still in his second year of undergraduate education, said that presenting his self-developed research in an international scientific environment is a valuable academic and personal experience. Stating that he has been developing research in the field of artificial intelligence for a long time, Bektaş used the expressions, “Throughout this entire process, my goal is not just to use existing technologies, but to understand the scientific mechanisms behind these technologies and develop new methods.”

Bektaş, evaluating the UbiComp/ISWC 2026 experience as an important step in his research journey, stated that he was happy to be part of an international scientific environment as a student of Üsküdar University. Bektaş expressed his hope that this experience would serve as a good example for himself and for other undergraduate students who wish to conduct research.