My Projects

The master has failed more times than the beginner has even tried. Here I proudly present my previous projects, no matter whether they were successful or not, for each of them helps me learn and grow.

Generating Uncommon Relations

boat_filled_with_water — Left: "boat filled with water"; Right: "vase made of flowers"

vase_made_of_flowers — Left: "boat filled with water"; Right: "vase made of flowers"

Text-to-image diffusion models are not good at composing objects with relations that are not common between them in the training data. During my internship at Geometric AI Lab in the summer of 2023, I proposed a method that finds relation token embeddings that can resolve this problem while using a pretrained diffusion model. While the project itself did not bring about significant technical advantages over the existing literature, it inspires more fundamental questions on how representations which describes object interactions or relations can be learned from data.

meowie — A NeRF trained on my kitty stress ball on Nerfstudio

As part of the internship, I also implemented PointNet, NeRF, and diffusion models in Pytorch. The code can be found here.

Robotic Arm System

arm_whole — Left: components of the system; Right: picking task

arm_pick — Left: components of the system; Right: picking task

As part of the coursework in Electronic Design Lab at KAIST in Spring 2023, I designed and implemented a 4-link manipulator robotic system with pick-and-place capability using C++. The system incorporated a BeagleBone Black processor, WiFi communication, a magnetic end effector, and a camera for teleoperation. Through the project, I learned a lot about inverse kinematics, motor control, and embedded systems.

**Implementation of Data Mining Algorithms [github] **

I implemented A-Priori, locality-sensitive hashing, collaborative filtering, PageRank, graph motif counting, and DGIM algorithms on real-world massive datasets using Apache Spark.

Interpretable Summarisation of Hypergraphs

tags-math-sx — Decomposition on Math Stack Exchange tag network

This project was for my Undergraduate Research Participation (URP) program at KAIST in Spring 2022. I formulated a new problem in hypergraph data mining and developed an information-theoretic algorithm. My practical contribution includes the implemention and evaluation various strategies for hypergraph clustering as a subroutine of the algorithm on 5 real-world datasets of different domains using SciPy, networkx, and the PyData stack.

The Coevolution of Topics and Emotions in News [report] [github]

This was my final project for the Computational Social Science course at KAIST in Fall 2021. We conducted topic modelling and sentiment analysis on a dataset consisting of 150,000 news article from 15 American publications using the PyData stack and scikit-learn. We proposed novel characterisations and measures for the evolution of topics and emotions in news based on Dirichlet Latent Allocation model.

back