Abdullah Hayran

• Project Galaxy: AI-powered scalable aerial path planning, in collaboration with Airbus • Took primary responsibility in research and development for the graph neural network solution, in PyTorch. • Iteratively refined the path planning methodology by researching and experimenting with various GNN architectures & methods in a fast-paced environment. • Delivered the final model with 2.7% average test error in terms of path lengths and 7x run-time improvement over traditional shortest path algorithms.

• Contributed to the research project BRAINet in VAVlab focusing on brain network classification, during junior and senior years. • Tackled the problem of insufficient and imbalanced medical data for machine learning applications. • Designed and implemented a brain network augmentation framework from scratch using NumPy, based on a variational autoencoder with label dependent prior, which improved classification accuracy by 7%.

• Designed and implemented convolutional neural networks for aerial image classification using Keras. • Utilized a low-volume and imbalanced dataset using transfer learning and image augmentation techniques, reached over 90% accuracy.

• Successfully designed and integrated a configurable digital low-pass filter into the analog-to-digital converter of a Xilinx FPGA board for high frequency noise suppression, using VHDL and MATLAB.

Completed the following projects using MATLAB: • Corner detection in hand-drawn figures • Real-time object tracking

Deep learning & computer vision