Junkeun Yi

I am a master's student at UC Berkeley, where I work on computer vision and machine learning.

Previously, I did my undergraduate studies at UC Berkeley, where I studied computer science.

Email  /  Resume  /  Github

Research

I'm interested in computer vision, especially in creating models that can reason about a scene through domains outside of its main downstream task. Currently, I am working on transformer models for video with the aim for publishing at a major AI conference.

Personal

I've lived in 6 different countries (Korea, New Zealand, Japan, Turkey, Pakistan and the US)! Also, I love spicy food!

Projects

PyTorch remiplementation of SAVi, an unsupervised / weakly supervised video model that learns representations of objects separate from each other and the background through slot attention. Original code by Google written in Jax.

PyTorch implementation of Policy Distillation with exploration by Random Network Distillation. The model learns to distill an expert PPO policy into a smaller student network while also taking exploration steps through trying to fit a randomly initialized network.

Python implementations of various computational photography algorithms

Java implementation of a SQL database with querying, indexing, joins, query optmization, concurrency, and recovery capabilities.

C implementation of an Operating System with multi-threading, scheduling, synchronization, and file system capabilities.

Education

Experience

Working on publication for video transformer model.

Implemented SAVi for PyTorch, and unsupervised video object understanding model.

Enlisted soldier (PV2 - SGT) in KATUSA program.

Performed English-Korean translation/interpretation and manned computing devices.

Open-Source contribution to the Greenplum Database (Postgres-based SQL database), adding backup utility and command-line interface features and fixing bugs.

Wrote pipeline for incremental changes between a database and its remote backup using Write-Ahead Log streaming, incorporating Kafka as a streaming medium and programmed consistency points for source-to-backup consistency. slides

Contributed in writing AWS Kubernetes elastic cluster contorller using Kubernetes Go-client, Metrics-Server client, and the Prometheus API for resource management.

website template from Jon Barron.