About Me

Hi, my name is Kimbo Chen, a MS CS student at University of California Riverside.

I have 2 academic goals:

• Making deep learning a robust and widely accessible technology.
• Accelerate the time and lower the cost of developing computer hardware.

I am active on Twitter (@kimbochen), where I write about deep learning, computer architecture, and things I learned in general.
You can also contact me using email: chentenghung at gmail dot com.

Projects

Experimenting with JAX

A non-trivial example of implementing a deep learning model using JAX.
Link to Twitter threads: Testing Phase | Full Project.

• It contains a minimal neural network library implemented from scratch using JAX.
• A small GPT model is implemented using the library, including the training and evaluating code.

PanoDPT

An implementation of the paper Dense Prediction Transformer.

• Refactored the original implementation and organized the project using PyTorch Lightning.
• Designed a 2D scatter operation using Torch Scatter to process panoramic images.

Systolic Array in PyRTL

Explained how a systolic array works using the register-transfer level design library PyRTL.
Link to Twitter thread

• Implemented a 2 by 2 systolic array using PyRTL.
• Walked through cycle by cycle how the systolic array computes a 2 by 2 matrix multiplication.

ALREC: Army Leave Report Chatbot

A chatbot I wrote when serving in the military to deal with the incoveniences of reporting status.
Link to Twitter thread

• Deployed a LINE chatbot on hosting service Heroku using Google Sheets as backend.
• Streamlined the reporting process of a 15-person message group.

Research Interest

I am interested in machine learning systems, MLOps, and computer architecture in general.
Here are some papers I read and summarized:

• Divide and Conquer: Leveraging Intermediate Feature Representations for Quantized Training of Neural Networks
• VOS: Learning What You Don’t Know by Virtual Outlier Synthesis
• Planaria: Dynamic Architecture Fission for Spatial Multi-Tenant Acceleration of Deep Neural Networks
• Dark Silicon and the End of Multi-core Scaling

I also covered many parts of the Tesla AI Day:

• Vision
• Auto Labeling
• Tesla Dojo Computer

Community Experiences

I attended many online events to meet other researchers and learn new things.
After attending, I wrote down what I learned. Here are some highlights:

• ICCV Undergraduates in Computer Vision
• Computer Architecture Long-term Mentoring
• Seminar Series on Tensor Computation - Halide
• Big Science Data Sourcing Sprint
• Machine Learning Tokyo Session - ResNet Strikes Back
• PyMC Open Source Sprint

Education

• B.A. in Computer Science, National Tsing Hua University
  • Sept. 2017 - Jun. 2021
  • Advisor: Hwann-Tzong Chen