Lead Data Scientist
Kite-balancing project (tech lead). Technical management of 2 researchers.
- Did initial research and development in algorithms behind kite-balancing
- Developed proof-of concept on application of GANs to sensory data augmentation (Keras, Tensorflow)
- Developed behavioral-cloning framework to learn kite balancing from demonstrations (of an operator), using deep learning (Keras, Tensorflow)
- Developed reinforcement-learning software to improve results of behavioral-cloning deep neural network using REINFORCE algorithm (and additional interactions with environment)
- Developed integration of AI solution into embedded system (controller programmable with Arduino-C language)
Tech stack: Python, Tensorflow, Keras, keras-rl, reinforcement-learning, gym