date
Jun 25, 2022
slug
resume
author
status
PublicOnDetail
tags
summary
type
Paper
thumbnail
updatedAt
May 11, 2024 04:29 PM
Mao PoYuan, Master Student of Artificial Intelligent and Data Science
Contact.
Email. mao.kyushu@gmail.com
Phone. (+81) 080 9649 6720
Channel.
Highlight.
- Publish a machine learning algorithm “Magnum” in journal of Neurocomputing (Impact factor 6).
- Project leader of the "Nanoscale 3D Printer", which has been awarded in two national competitions.
- 3 years + of experience in deep learning and machine learning.
- 2 years of experience in backend development. (Work: MiraiX company website)
Experience.
Master Degree - Research in Image Generative Diffusion Model • Laboratory of Intelligent Systems ( Kyushu University)
2022.10 - present
Description
Researching in diffusion based model, such as Imagen, Stable Diffusion, DalleE and GLIDE.
What did I do
- Implemented model's pipeline (Stable Diffusion, CLIP) from scratch using PyTorch, including building a custom dataloader and pipeline, and modifying the model's formula for fine-tuning.
- Deployed the model's observation visualization dashboard interface on a local Linux server.
Tech Stack
Machine Learning Library: Tensroflow, PyTorch, Pandas, Numpy
Web Scraping Library: Beautifulsoup, Selenium, re
Visualization: Plotly dashboard
Result
• Under review “Synthetic Shifts to Initial Seed Vector exposes the Brittle Nature of Latent-Based Diffusion Models”
Under review “HOMOE: A Memory-Based and Composition-Aware Framework for Zero-Shot Learning with Hopfield Network and Soft Mixture of Expert ”
Research Assistant - Backend Development • Laboratory of Intelligent Systems ( Kyushu University)
2022.10 - present
Description
Development of a python and machine learning teaching website for MiraiX company.
What did I do
Backend design and management (Design note).
Created an authentication system and a database system using Firebase.
Created a payment system using Stripe.
Tech Stack
Firebase, Javascript, Stripe
Research Student - Research in Bioinspired Unsupervised Learning Algorithm • Laboratory of Intelligent Systems ( Kyushu University)
2021.09 - 2022.09
Description
Research in cognition-inspired model. Focused on how to utilize chunking strategy inside the machine learning models.
What did I do
- Migrated spike MRIL loss onto Recurrent Neural Networks Neural Networks.
- Enhanced the scalability of SyncMap.
Tech Stack
Machine Learning Library: Tensroflow, PyTorch, Pandas, Numpy
Visualization: Plotly dashboard
Result
Neurocomputing Volume 550 “Tackling High-Dimensional Structures with Self-Organization”
SICE2021 “Preliminary Results on Chunking with Recurrent Neural Networks”
Research in Adversarial Sample
2020.12 - 2021.08
Description
Carried out adversarial attack on Resnet50 to test the model’s robustness.
What did I do
- Implementing attack algorithms such as FGSM, One-pixel attack, AdvGan, and Deepfool from scratch.
- Verified Resnet50 facing different adversarial sample generative algorithm.
Tech Stack
Tensorflow, Matplotlib
Mao718/attack_zoo
Backend Engineering
2021.04 - 2021.08
Description
Setup the database locally and built secure authentication and others API connect with the database.
What did I do
Create API interacting with Postgres SQL.
Tech Stack
Fastapi, SQL, linux
2018.09 - 2020.01
Description
Bring 3D printer into nanoscale is not only expensive but also complicated. Our target was to create a cost-efficient nanoscale 3D printer.
To lower the costs, we utilized the laser inside a CD player. And we created a first-order control system that receives the light sensor signal for the laser and controls the three-axis step motors and the laser motor.
What did I do
- Designed and built up the first-order electronic control system.
- Implemented motor control program of Arduino with C++.
- Create a human interface using C#.
Result
- Awarded the Best Innovation Award in the Mechanical and Automation Category of the "2019 National University Industry and Academic Innovation Competition".
- Awarded "2019 Taiwan Precision Engineering Technology Symposium Hanmin University Project Competition".
Tech Stack
SolidWorks, Arduino(C++), C#
Language:
Chinese - native
English - business level
Toeic : 805
Japanese - conversational level
JLPT : N1