Here is my resume (last updated 11/24).

Professional

Education

PhD in Applied Mathematics at UCLA. (2019-2024)
- Thesis: Novel Applications of Neural Networks in Physics-Based Simulations
- My research interests lies in between computational linear algebra and machine learning. In general, main focus of my projects was to improve existing algorithms/methods that are used in physics-based simulations via learning-based approaches.

BS in Mathematics of Computation at UCLA. (2015-19)
- Sole recepient of Math Undergraduate Merit Scholarship in 2015
- Graduated Cum Laude

Research Intern at Epic Games (6/2022 - 12/2024).
- Designed a novel learning-based method to compute signed distance function (SDF) of human body motion depending on the rotation of joints in the skeleton, so that the deformation of the skin/muscle (especially near the joints) looks as real as possible.
- Our method, called ML-SDF, is a combination of multiple ML models in which each represents different parts of the human body that involves a joint (e.g. elbow, shoulder, spine, legs, etc.).
- ML-SDF is used for character animation and its interaction with clothing in Unreal Engine. The method reduces the memory cost on the RAM significantly (less than 1% memory requirement compared to the actual levelsets it's replacing). The goal of the project is to automatize the whole process given the body mesh and its skeleton (e.g. different characters of Fortnite). This includes joint-based body part selection, dataset generation for the corresponding body parts, and the training, all to be done inside Unreal Engine.

A Deep Conjugate Direction Method for Iteratively Solving Linear Systems (2021 Feb - 22 Summer)
- Created CNN-based linear system solver, a modified version of the classical Conjugate Gradients (CG) method, for the systems coming from computational fluid dynamics applications (e.g. incompressible flows).
- Designed the network architecture and its custom loss function to perform unsupervised learning to speed up the solver. At the moment DCDM is 3.2 times faster than classical methods for the linear systems over 56 millions of degrees of freedom.
- Developped a novel method for creating the dataset so that it does not require any simulation input, as opposed to our competitors. This reduced the dataset size significantly, and consequently the time required to generate it and the training.
- The work is accepted and presented at ICML 2023 at Hawaii. Here is presentation.

Research Intern at UCLA IPAM RIPS GumGum Team. Here is resulting paper: [arXiv] or [springer], and poster (18 Summer).
- Worked on a research team that build automated video augmentation pipeline that identifies crowd regions in sport videos and overlays an advertisement onto these regions, so that the advertisement looks aesthetic and natural for human perspective. This video link is an example of the final product.
- Designed and implemented an algorithm to visualize RGB-Depth image frame in 3D. Implemented Random Sample Consensus Algorithm in this 3D-reconstruction to detect planes to overlay advertisement onto the image in 2D.
- Developed a novel metric on video frames for the evaluation of the PSPNet crowd segmentation algorithm.

Undergraduate Researcher at Computational and Applied Mathematics REU. Here is resulting paper preprint [arXiv] (17 Summer).
- Worked on a research team that applied image processing techniques to extract non-content-based features from LAPD body-worn-camera video recordings and used machine learning to classify the recordings into categories.
- Created color-based feature and its metric system for video frames and implemented in the ML Algorithm.
- Modified the old machine learning model to accept the new dataset of LAPD videos. Experimented with tuning hyperparameters and reduced the error by 50 percent.

In College
- Putnam Performance Prize by UCLA Mathematics Department, 2019
- Honorable Mention at William Lowell Putnam Mathematical Computation, 2018
- Richard F. Arens Putnam Scholars Undergraduate Award, 2016
- Sole recipient of UCLA Math Undergraduate Merit Scholarship in class of '19. Since its foundation at 2010, there are only 8 recipients of this scholarship all over the world.