Minjae Kim

TL;DR: a Ph.D. student who is interested in deep learning system optimization for mobile devices.

I am a Ph.D. student in the department of computer science and engineering at Seoul National University. I received master’s degree in computer science from SNU and my bachelor degree in computer science from POSTECH.

I am currently in Human-centered Computer Systems Lab (advised by Prof. Youngki Lee), was in Software Platform Lab (advised by Prof. Byung-Gon Chun), and worked as an undergraduate research intern at Computing Systems Lab (advised by Prof. Hyojin Sung).

Now, I am interested in accelerating computation heavy systems with software optimization for heterogeneous mobile devices. Currently, my main focus is on deep learning but I don’t limit my interest to it. I strongly believe that cross-domain optimization from silicon to human is essential for deep learning (or any other areas) to penetrate our daily lives.

Publications

• Jingyu Lee, Hyunsoo Kim, Minjae Kim, Byung-Gon Chun, Youngki Lee. Maestro: The Analysis-Simulation Integrated Framework for Mixed Reality, The 22nd ACM International Conference on Mobile Systems, Applications, and Services (MobiSys’24) (To appear)
• Seonjun Kim, Minjae Kim, Youngki Lee, A Joint Analysis of Input Resolution and Quantization Precision in Deep Learning, Proceedings of the 29th Annual International Conference on Mobile Computing and Networking (MobiCom’23) (Poster) (Link)
• Hyungkyu Ham*, Hyunuk Cho*, Minjae Kim, Jueon Park, Jeongmin Hong, Hyojin Sung, Eunhyeok Park, Euicheol Lim, Gwangsun Kim. Near-Data Processing in Memory Expander for DNN Acceleration on GPUs, IEEE Computer Architecture Letters (CAL) [*: Equal contribution], Nov 21. (Link)

Experiences

• Machine Learning Engineer @ Team ISLAND, Aug 2020 ~ Sep 2020
  Worked on building a lightweight 3D pose estimation model via knowledge distillation. To distill the knowledge from the regression-based teacher model, conducted label smoothing based on a gaussian filter to the output of the teacher model and jointly trained the student model with both the teacher model and the ground truth.
• Software Engineer (Outsourced work) @ Richbeam, May 2020 ~ July 2020
  Migrated the complex backend system consisting of Flask, Tornado, and gRPC servers, which was already deployed on the private on-premise server, to the AWS Cloud with load balancing techniques.
• Software Engineer (Full-time) @ Cryptoquant, Sep 2019 ~ Mar 2020
  Worked on the whole infrastructure used for on-chain analysis and data provisioning. The system involves PostgreSQL, Flask, Redis, and MongoDB. 99.997% of the availability was achieved in the designed system, which is crucial in the financial field. Also, participated in an anti-money laundering system, which helped to solve the biggest sexual abuse “N-th room” in Korea.
• CTO & Co-Founder @ DSigner, Sep 2018 ~ Sep 2019
  Developed a font navigation system to help designers to explore the proper font in a huge amount of typography. Auto-encoder-based embedding was engineered with the expert knowledge of typeface. Also, worked on backend servers using Flask, MySQL, and MongoDB. Interviewed in YC 2020 Spring Batch.
• Research Intern @ HodooAI, Apr 2018 ~ Aug 2018
  Researched transferring makeup between non-labeled face images. By matching the corresponding regions with local embeddings and applying the color transfer techniques on those matches, successfully transferred makeup without any guidance from humans.
• Software Engineer Intern @ byclip, Jun 2017 ~ Aug 2017
  Developed a chatbot management system with MongoDB and Express, and developed an interactive item-based recommender system for restaurants.

Others

I was a finalist of International Science and Engineering Fair sponsored by Intel (Intel ISEF 2015).

I deejay house music, especially for deep house and tech house.

If you are interested in working with me, please contact me via email widiba03304 at gmail dot com. When you really want to contact me, please add the prefix [HELLO] to your title to ensure that you read this.