hello

Hello!

I’m Jiachen Jiang

Signal Processing & Machine Learning Engineer

Hello! I’m Jiachen Jiang.

I'm a master student of Electrical and Computer Engineering (ECE) at University of Michigan - Ann Arbor. My track is Signal Processing & Machine Learning.

My research interest includes Image Processing, Computer Vision(CV), Natural Language Processing(NLP) and Optimization method. Currently, I'm working with Prof.Qing Qu on Neural Collapse and Deep Image Prior.

Age:
24
Email:
jiachenj@umich.edu
Wechat:
j1146904763
Phone:
+1 734-846-5481
Address:
2552 Stone Rd, Ann Arbor, MI, US
Staus:
Available
Walter Patterson

Education

BeiHang University

Bachelor at Beijing, China (2016/09-2020/07)
ui-ux

Major: Electronic Engineering; School of Engineering; Rank: 9/293; GPA: 3.82/4.00

Honors:

  • Second Prize in National Undergraduate Electronics Design Contest, 08/2019;
  • First Prize in Beijing Integrated Circuit Design Competition, 06/2019;
  • Group Silver Award of Youth Entrepreneurship Competition, 06/2018;
  • Second Prize of Mathematical Modeling Competition, 05/2018;
  • Third Prize of Fengru Cup Creative Thesis Competition, 09/2017;

University of Michigan

Master at Ann Arbor, MI (2020/09-2022/06)
app development

Major: Electronic and Computer Engineering; Rackham Graduate School; GPA: 3.99/4.00

Coursework:

  • Probability and Random Process(EECS 501)
  • Foundation of Computer Vision(EECS 504)(A+)
  • Matrix Method for Signal Processing(EECS 551)
  • Machine Learning(EECS 545)
  • Image Processing(EECS 556)
  • Optimization Method(EECS 559)
  • Intermediate Python Programming(SI 507)
  • Self-Driving Cars(ROB 535)
  • Natural Language Processing(EECS 595)(A+)
  • Computer Architecture(EECS 595)

Internship

Institute of Automation, Chinese Academy of Sciences

Beijing, China (2019/07-2019/10)
ui-ux

Main research direction: Concentration level detection based on EEG signal processing;

Contribution:

  • In charge of most of the programming tasks,including model building, model refinement, algorithm implementation, and debugging.
  • Wrote the summary part in the patent writing and described the technical route of the algorithm.

University of Michigan

Ann Arbor, MI (2022/06-2023/06)
app development

Main Research Directions: low-dimensional model from high-dimensional space and numerical optimization method.

Specific Directions:

  • Neural Collapse
  • Deep Image Prior

Projects

Hyperspectral Images Denoising via a Tensor Dictionary

EECS 556 Winter 2021
ui-ux

As 3rd order tensors, hyperspectral images (HSIs) can deliver more authentic representations for real scenes, enhancing the performance of many computer vision tasks when compared with traditional RGB or gray-scale images.

In this paper we propose an effective HSI denoising approach based on the Tensor Dictionary Model by considering two intrinsic characteristics underlying an HSI: the nonlocal similarity over space and the global correlation across spectrum. We modify the original method of grouping similar patches into clusters and optimizing over each cluster by designing an optimization problem with an objective function targeted at representing the HSI with a sparse global dictionary representation. Iteration over variables is used to solve the nonconvex Lagrange dual problem.

Monitoring Social Distancing and Mask Wearing with Person Detection and Tracking

EECS 504 Fall 2020
ui-ux

Keeping proper social distance and wearing masks is probably the best way to prevent the spread of the COVID-19. We created a unified pipeline for social distance and mask wearing monitoring for real-time video. We highlight human localization, homography, face detection and mask detection.

Main Method:

  • Human and Face Detection: Mask-RCNN human instance detector.
  • Mask Detection: Binary classification with an image up-sampler, a feature extractor and a classifier.
  • Social Distance Highlighting: Highlighting a 6-feet social distance contour for persons in the video.

A City Search Engine Website Design Based on Flask

SI 507 Fall 2021
ui-ux

This project can search the top 5 cities in population in selected country and get real time Twitter related to the cities. The user would select a country they interest and choose the sort types. Then they would get a table containing the information about the cities, including WikiData Id, Latitude, Longitude, Population and Distance from Ann Arbor. They can choose sort the table by population or by the distance from Ann Arbor.

Main API:

  • GeoDB Cities API: Provides basic information about cities, counties, regions, and countries throughout the world.
  • Twitter API v2.0: The Tweets Look up API is one of the primary resources on Twitter.We look up tweets related to the city on it.

Racing on a Pre-Defined Map with Unknown Obstacles with PID Controller

ROB 535 Fall 2021
ui-ux

The control project of controlling a bicycle model mainly contains two parts as follows.

  • Task 1: We are required to design a controller for the system to get from the beginning to the end of a pre-defined track as rapidly as possible;
  • Task 2: We need to develop a control design algorithm (which may or may not modify the controller constructed in the first task) to avoid obstacles that are known only at run-time.

I mainly focus on task2. There are three parts of my work, including Avoid obstacles module, generate reference speed and angle module and PI controller module.

Dual Multi-head Co-Attention For Abstract Meaning Reading Comprehension

EECS 595 Fall 2021
ui-ux

We work on SemEval-2021 shared task 4, which requires the participating system to fill in the the correct answer from five candidates of abstract concepts in a cloze-style to replace the @Placeholder in the question.

In this work, we basically choose ELECTRA as our encoder, and try to add Multihead Attention Multichoice Classifier(MAMC) and DUal Multi-head Co-Attention (DUMA) classifier as a one-layer classifier. They both achieve higher performance than ELECTRA itself. On conclusion, our ELECTRA + DUMA approach tends to perform out other methods as our best result, it ranks 3rd for task 1 and 5th for task 2 with the accuracy of 89.95%, 91.41%.

A 3-way Superscalar Out-of-Order CPU with P6 style Register Renaming

EECS 470 Winter 2022
ui-ux

We implemented a P6 style register renaming architecture CPU in Verilog.

Our processor is designed to have 6 stages in total: Fetch, Dispatch, Issue, Execute, Complete, Retire.

  • Fetch Stage: Fetch at most 3 instruction in order.
  • Dispatch Stage: Decode the information of the instruction and read the register value from register file.
  • Issue Stage: Instructions in RS with zero T1/T2 would request to issue.
  • Execute Stage: 3 ALUs with 1 cycle latency and 3 8-stage multiplier.
  • Complete Stage: Load/store instructions would go to the in-order LSQ and wait for the valid data from Dcache.
  • Retire Stage: The head of the ROB would be written into register file.

My Skills

I am a quick learner and specialize in multitude of skills required for Signal Processing & Machine Learning.

Python70%
Matlab90%
Julia80%
Pytorch90%
Machine Learning85%
Deep Learning90%
Computer Vision80%
Natural Language Processing80%

Contact Me

Let’s talk how I can help you!

If you like my work and want to avail my services then drop me a message using the contact form.

Or get in touch using my email or my contact number.

See you!

Email:
jiachenj@umich.edu
Wechat:
j1146904763
Phone:
+1 734-846-5481