Projects
Some research / academic projects I've worked on:
π
Sep. 2018 - May 2022 β’ πMore Info β’ π CVPR Paper β’ π Main Website
A drone flight system to collect over 120,000 images. The dataset was used to conduct experiments showing severe vulnerabilities in neural networks to pose & camera shake, which we improve by 32%. Findings published to CVPR 2019.

π
Jan. 2021 - May 2022 β’ πMore Info β’ π CVPRW Paper β’ π Main Website
A novel neural network refinement algorithm to generate 3D meshes from a single image. Used self-supervised learning & symmetry regularization; beats state-of-the-art (up to 47%), across many datasets.

π
Feb. 2021 - May 2022 β’ πMore Info
A neural network model currently in development for self-driving cars. Utilizes multimodal fusion with 2D images and 3D pointclouds to predict navigation actions. Additionally, 2D and 3D explanations are jointly predicted to provide intuitive explanations for model decisions.

π
Feb. 2021 - Jul. 2021 β’ πMore Info β’ π° Slides
An interactive dashboard to analyze user chat logs and describe their linguistic behavior. NLP transformer models (RoBERTa, GPT-2) are utilized for sentiment analysis, clustering, style transfer, & generative modeling.

π
Jun. 2020 - Dec. 2020 β’ πMore Info β’ π Paper
Studies the application of several domain adaptation techniques (MMD, Deep CORAL, DANN) to enable unsupervised transfer from CAD ShapeNet data to real world data, for the task of single view 3D reconstruction.

π
Jul. 2020 - Nov. 2020 β’ πMore Info β’ π Report
Proposes a unified taxonomy to generalize most of the current methods in unsupervised domain adaptation. Also takes a deep look into Contrastive Adaptation Networks, and ways to improve performance.

π
Mar. 2020 - Jun. 2020 β’ πMore Info β’ π Report
AI that learns to play Connect Four, with reinforcement learning (Q-Learning & Monte-Carlo policy iteration).

π
Jun. 2015 - Sep. 2016 β’ πMore Info
A suite of tools to analyze image sensor quality. Depicts the sensorsβ silicon wafer with indications of areas with the most unstable sensor chips. Also provides functions like history recording, re/undo, statistics calculation, pixel editing, and image transformations, behind a user-friendly GUI.
