Jiayi (Tris) Tian
Email:
jiayi_tian@ucsb.edu | Mobile: 805-2450298
Education
University of California, Santa Barbara Santa
Barbara, CA
Department of Electrical and Computer Engineering
Sept.
2023 - ongoing
■
MS/PhD, Major in Computer Engineering
Nanjing University Nanjing,
China
School of Electronic Science and Engineering
Sept.
2019 - Jul. 2023
■
B.Eng., Major in VLSI Design & System
Integration
■
Cumulative GPA: 4.51/5.0; Major GPA:
4.49/5.0 (Top 10% in the grade)
Relevant Coursework: Digital
System I(96), Experiments in Digital Logical Circuits(96),
Design of Deep Learning System Based on Hardware Acceleration of FPGA(92),
Fundamentals of IC Design(83), VLSI DSP Architectures(89), Digital Logic and
Intelligent Processor Design(87), IC design and solution(91.6), SoC Design
Methodology(94)
Technical Skills
■
Programming Languages: C/C++, Python/Pytorch, Verilog, Matlab, LaTex
■
Hardware Design and Simulation Tools: Vivado/Vitis, Quartus, Modelsim,
Design Compiler, VCS, DVE, Altium Designer, Multisim, SPICE
■
Development Tools: VSCode,
Git/Github, Bash, Jupyter
Notebook
Publications
■
Jiayi Tian, Chao Fang, Haonan Wang and Zhongfeng Wang.
"BEBERT: Efficient and robust binary ensemble BERT." IEEE
Conference on Acoustics, Speech, and Signal Processing (ICASSP). 2023 [accepted]
Experience
Zhang’s Group, University of California,
Santa Barbara Santa Barbara, CA
Graduate Student Researcher, collaborated with Intel's Component Research June. 2023 - ongoing
■
Constructed an end-to-end tensorized
Transformer training FPGA accelerator with software and hardware co-design.
■
Designed a hardware-friendly end-to-end
tensorized Transformer training framework with 31x compression ratio without
performance degradation using Python and Pytorch.
■
Developed a synthesizable tensorized
Transformer training engine using C++, including forward propagation, backward
propagation, gradients computing and weight updates.
■
Devised efficient dataflows for tensorized
Transformer training with parallel and pipeline using Vitis HLS.
ICAIS Lab, Nanjing University Nanjing,
China
Research
Assistant. Apr. 2021- Oct.
2022
■
Proposed Binary Ensemble BERT (BEBERT), a
novel compression scheme to boost the training efficiency, model accuracy, and
robustness of binary BERT for accurate and hardware-friendly inference.
■
Developed a binary ensemble BERT
finetuning algorithm to construct highly compressed BERT with 13x reduction in
model size and 15x reduction in computation cost using Python and PyTorch.
■
Published "BEBERT: Efficient and
robust binary ensemble BERT" on ICASSP23.
projects
ICAIS Lab, Nanjing University Nanjing, China
Undergraduate Graduation Project.
Nov. 2022- June. 2023
■
Developed an early-exit BEBERT for
efficient and accurate inference design.
■
Accomplished an ensemble algorithm with
the early-exit mechanism for binary BERT models, further reducing the
computation cost by 20%~40% compared to BEBERT in the GLUE benchmark.
■
Used Python and Pytorch
to rewrite critical operations in binary BERTs for hardware accelerator design.
ICAIS Lab, Nanjing University
Sept. 2021- Sept. 2022
Team
Project.
■
Designed an INT8 quantized BERT model for
hardware-friendly inference using Python and PyTorch.
■
Used Matlab to
achieve critical operations (Attention, Softmax.
etc.) for hardware design.
VLSI
Design Course Project Mar. 2022- Jun. 2022
Individual assignment, A (top 5%)
■
Used Synopsys's VCS, DVE and Design
Compiler tools to devise, evaluate and synthesize efficient one-dimension
convolution computing.
■
Proposed three optimization methods based
on basic VLSI techniques, including pipeline, parallel, and transpose, to
accelerate the convolution process and reduce the area.
Verilog Design Course Project Mar. 2021- Jun.
2021
Team
Project, A+
■
Used Quartus and Intel Cyclone5 Series'
FPGA to complete a VGA display clock on the monitor, which could set up time by
typing on the keyboard.
■
Analyzed the temporal logic of the VGA
interface to achieve the correct time sequence and used RAM to store the
address for the hands of the clock displayed on the screen.
Honors and Awards
■ People's
Scholarship, The 2nd Prize in NJU, Dec. 2022 (10%)
■ Excellent
Student, NJU, Dec. 2022(5%)
■
Huawei "Intelligent Base"
Scholarship, Dec. 2022 (<1%)
■ Excellent
Volunteer Prize on the school's 120th anniversary, Sept. 2022 (<1%)
■ Excellent
Volunteer Prize, NJU, Dec. 2021 (<1%)
■
Jinxiao
Company Scholarship, Nov. 2021 (5%)
■ People's
Scholarship, The Academic Competition Award, Nov. 2021 (5%)
■ National
Undergraduate Electronic Design Contest, The 2nd Prize in Jiangsu Province,
Nov. 2021 (30%)
■ Excellent
Department Director, Student Union in Sch of Elec Sci and Eng., NJU, Oct. 2021
(15%)
■ People's
Scholarship, The 2nd Prize in NJU, Nov. 2020 (10%)
■ National
Undergraduate Electronic Design Contest, The 2nd Prize in Jiangsu Province,
Oct. 2020 (30%)