Pongsakorn U‑chupala

Research Engineer

Profile

Just a research engineer with an intense passion for modern Japanese sub-culture including anime, manga, and games.

Education

Doctor of Engineering, Computer Science
Master of Engineering, Computer Science
Graduate School of Information Science, Nara Institute of Science and Technology
Bachelor of Engineering, Computer Engineering
Department of Computer Engineering, Faculty of Engineering, Kasetsart University

Roles

When What
2018‑present Research Engineer, Distributed Machine Learning
R&D Center, Sony Corporation
  • To enable engineers and data scientist to utilize standard AI/ML toolchain on non-standard proprietary GPGPU cluster, I developed custom distributed deep learning stack including custom collective communication solution for non-standard proprietary communication fabric.
  • I help design GPGPU cluster using non-standard proprietary hardware for distributed deep learning workload.
  • Using low-rank learning method we developed in collaboration with UW-Madison, I reduced memory footprint of several of our neural network learning tasks by up to 50%.
  • I coordinate our team research collaboration effort with UW-Madison, which resulted in the publication: PUFFERFISH: Communication-efficient Models at No Extra Cost.
  • I coordinate our team research collaboration effort with Georgia Tech, which resulted in the publication: Nested Dithered Quantization for Communication Reduction in Distributed Training.
  • I developed a distributed deep learning simulator, which help our team broke world record of ImageNet/ResNet-50 training speed. During ABCI Grand Challenge 2018, we gained access to the entire ABCI cluster only for a limited time. The simulator allows us to do dry hyper-parameter tuning, thus significantly reducing the number of experiments required on the cluster.
  • I work on NNabla, Sony’s high-performance deep learning framework. I am responsible for distributed learning performance optimization as well as designing next-generation distributed learning API.
2015‑2018 Ph.D. Student, Computer Science
Nara Institute of Science and Technology
  • Doctoral Dissertation Increasing Data Center Efficiency with Improved Task Scheduling and Communication I propose several optimizations for cloud infrastructure.
  • Master’s Thesis Overseer: Application-Aware Routing OpenFlow controller for bandwidth and latency aware routing implemented with POX.
  • PRAGMA-ENT Breakable international SDN testbed for PRAGMA community. I help established and maintained this network, which connect multiple institutions including NAIST, Osaka University, University of California San Diego, and University of Florida.
  • Applying Deep Learning to Network Traffic Identification and Categorization I developed network traffic classification model using stacked denoising autoencoder in TensorFlow. This model is learned on the CAIDA Internet traffic dataset. The model is a part of my proposal to create automatic SDN-based data center network traffic optimizer.
  • Container Rebalancing I proposed a novel scheduling mechanism with a rebalancing processing working alongside a scheduling process. A Hadoop/Hive-powered data processing technique and a Python-based simulation using Google’s cluster data is performed to validate this method.
2017 Intern Researcher
Information Technology Research Institute, AIST
I was responsible for deploying and benchmarking an experimental multi-site GPFS cluster connecting Japan, Australia, and U.S.A. The work involves the administration and debugging of Linux environment as well as collaborating with researchers from multiple institutions.
2014 Visiting Scholar
CalIT2, University of California San Diego
2013 External Researcher
HPCNC Lab, Kasetsart University
Developer
Innovative Extremist Co., Ltd.
2012 Google Student Ambassador SEA
Google Inc.
Outsourced Developer
Onebit Matter Co. Ltd.
Outsourced Developer
Diversition Co., Ltd.
2011 Exchange Student
FrontierLab@OsakaU, Osaka University
2009‑2010 Part-time Developer
Thoth Media Co., Ltd. (now Thoth Zocial Co., Ltd.)

Scholarships and Research Grants

  • MEXT Scholarship (2013-2018)
    • Full scholarship for Master program and extended to Doctoral program at NAIST
  • CICP2016 (2016)
    • Research grant for the project: Applying Deep Learning to Network Traffic Identification and Categorization
  • CICP2013 (2013)
    • Research grant for the project: eCOStamp
  • JASSO Scholarship (2011)
    • Scholarship for FrontierLab@OsakaU2011 student exchange program

Affiliations

Languages

  • Thai: Native
  • English: Full professional proficiency
    • TOEIC IP: 990 Points (June 2015)
    • TOEFL iBT: 101 Points (October 2012)
  • Japanese: Limited working proficiency
    • JLPT: N2 (September 2021)
  • Programming Languages: Python (expert), C/C++ (proficient), JavaScript (prior experience), PHP (prior experience)

Professional Experiences

  • Distributed Machine Learning (PyTorch, TensorFlow, NNabla)
  • GPGPU Application Optimization (CUDA, ROCm)
  • Distributed Communication Algorithm
  • High-performance Computing for Machine Learning Workload
  • Cloud Computing, Virtualization, and Linux Containers
  • Linux Administration
  • Software-Defined Network and OpenFlow
  • Software Engineering

Interests

  • Computer Networking Technology
  • Web Technology and Web Development
  • Smart Home Technology
  • Open Source Software

Contact

Resume (PDF)

Photo