TA for STAT 5330: Graduate Stochastic Processes

TA for STAT 4330: Undergraduate Stochastic Processes

• Proposed a few-shot conditional diffusion model for predicting vertical wind velocity on the globe to improve on NASA Goddard existing model.• Developed an LSTM model for predicting vertical wind velocity by taking into account time-dependent spatial information on the entire globe, which achieved 6% higher accuracy, while taking 8 hours less time to train than NASA model Wnet.• Developed new data-feeding methodology, which improved on MSE and cut the training time in half.• Presented the results in front of the Institute of Pure and Applied Mathematics (IPAM) staff, NSF foundation, and NASA scientists

• Collaborated with the Head of Swell Research to develop the first official Automatic Market Maker of Pendle tokens.• Developed statistical arbitrage model to execute pair trades of SUI and AVAX on the blockchain with a 4% profit.• Programmed Barra Risk Factor and Multiple Pairs Trade models, which achieved Sharpe ratio of 2.2 and 1.8 on the 2020-2022 data respectively.

• Created an algorithm for identifying monarchs in space with 98% accuracy, using QR codes with Hamming code.• Performed FFT and regression on data to determine whether M3 Bio logger attached to a monarch butterfly affects its stamina.• Developed a pipeline for 95% accurate object-recognizing AI based on the YOLO neural network

• Led a team of 3 students to develop an emotion-recognising RL algorithms for EEG.• Collaborated with CEOs of Curia and Fusion Health to develop the pipeline that recognises anger patterns with 94.7% accuracy, which is comparative to state-of-the-art results.• Explored different signal cleaning techniques under the guidance of a visiting Princeton math professor, Ovidiu Calin