新竹港式料理.jpg

Rocking Lai

Introduction 

I am a deep learning engineer as well as professional software engineer. I take part in various deep learning & computer vision projects, from software application to AI chip. Also have the experience of developing video surveillance system (full-stack). In addition, I strive for learning and constantly looking for ways to share my knowledge. I am the mentor of Coursera, which helps students in learning computer vision and machine learning.

Skills 

Domain Knowledge 

  • GPU Programming
  • Deep Learning Compiler
  • Neural Network Quantization
  • Computer Vision Algorithm
  • Surveillance System

Tool 

  • C++
  • Python
  • Pytorch
  • OpenCV

Experience 

Machine Learning Software Engineer, AMD, Sep 2021 - Present 

  • Develop GPU backend of various deep learning framework. Eg. PyTorch, Tensorflow, Onnx runtime
    a. Implement and optimize GPU kernel
    b. Proposed complex fused kernel, such as gemm + layernorm, gemm + reduction, gemm + softmax.
    c. Low precision arithmetic for deep learning.

Deep Learning Engineer, CVITEK, Oct 2019 - Sep 2021 

  • Spin off from Bitmain.
  • Mainly focus on deep learning accelerator (so called NPU or TPU) and stereo vision accelerator.
  • Deep learning accelerator
    a. Research and prototype the int8, int4 and bf16 quantization algorithm.
    b. Mix-precision algorithm based on Post training, result had applied patent in US and CN (US20220129736A1, CN114492721A).
    c. Model compression flow (heterogeneous quantization).
    d. Design quantization flow in deep learning compiler (graph optimization, calibration, fine tuning, mix-precision).
    e. Implement frontend of deep learning compiler (high level optimization and lowering to low level IR).
    f. Inference simulator via high level IR.
    g. Co-work with IC designer to design the cmodel of AI accelerator.
  • Stereo Vision accelerator
    a. Prototype the hardware-friendly stereo matching algorithm.
    b. Co-work with IC designer to design the cmodel.

Imlxnw96k4pw5ecmasom

Deep Learning Engineer, Bitmain, Aug 2018 - Oct 2019

  • Mainly focus on deep learning accelerator (so called NPU or TPU and so on)
  • Research of deep learning algorithm for edge AI accelerator.
    a. int8 and bf16 quantization algorithm.
    b. Post training based mix-precision algorithm.
    c. Quantization-aware training flow.
  • Deep learning compiler
    a. Quantization tool (calibration, fine tuning).
    b. Inference simulator for high level IR (cpu & gpu).
    c. Co-work with IC designer to design the cmodel of AI accelerator.

Imlxnw96k4pw5ecmasom

Algorithm Engineer, ULSee, Jun 2017 ~ Aug 2018

  • Vision algorithm for robot (object detection, gesture recognition, posture recognition).
  • Facial landmark tracking algorithm.
  • Improve face recognition flow.
  • Driver fatigue detection, phone talking detection for ADAS (Advance driver assistance system).
  • Plan & design the face recognition system for various projects. (IP camera integration, video management, lead the scrum flow).

Ovuysyw4opw53isngtk5

Software Engineer, NUUO, Sep 2014 ~ Oct 2017

  • Design and maintained the Network video recorder (NVR). It is an embedded Linux, which can received video stream from IP camera, various type of recording, video analytic, third party integration.
  • Develop new features for NVR client, which can play live video, playback video, smart search video, event management...etc.
  • Develop SDK of NVR, provide a way for third party to integrate our NVR.

Xpq7pojdp65cn0wbuqnj

Volunteer 

Course Mentor, Jan 2017 - Feb 2019

  • Machine Learning, offered by Stanford University.

  • Fundamentals of Digital Image and Video Processing, offered by Northwestern University.

Education 

Master's degree, Computer Science and Information Engineering

National Chiao Tung University (2012 - 2014)

Bachelor's degree, Computer Science

National Chiao Tung University (2008 - 2012)

Publication 

  • Toward Community Sensing of Road Anomalies Using Monocular Vision, IEEE Sensors Journal (Volume:16 , Issue: 8 ) 2016
  • Vision-Based Road Bump Detection Using a Front-Mounted Car Camcorder, IEEE International Conference on Pattern Recognition (ICPR 2014)

Patent 

  • Mixed-precision quantization method for neural network, US20220129736A1 (2021)
  • 神經網路的混合精度量化方法, CN114492721A (2020)
Powered By
CakeResume