Sheng-Yen Chou
email: sc3379 [at] cornell.edu

I am a research asistant at The Chinese University of Hong Kong (CUHK). I am interested in physic-informed machine learning, trustworthy AI, and reinforcement learning (sometimes plays with distributed system). My goal is to build a reliable AI system that can be impact the real world massively.

I received Bachelor's degree in Computer Science from National Tsing Hua University (NTHU). Before graduate, I was also a research assistant at DataLab and advised by Prof. Shan-Hung Wu. It's my pleasure to work with Dr. Pin-Yu Chen and Prof. Tsung-Yi Ho after graduate.

sym

Cornell University
Incomming Ph.D. in CS
Sep. 23 - Present

sym

Carnegie Mellon University
Safe AI Lab, Research Internship
Jun. 23 - Oct. 23

sym

The Chinese University Hong Kong
Research Assistant
Jul. 22 - Aug. 24

sym

National Tsing Hua University
B.S. in CS
Sep. 17 - Jun. 22

  Publications
sym

VillanDiffusion: A Unified Backdoor Attack Framework for Diffusion Models
Sheng-Yen Chou, Pin-Yu Chen, Tsung-Yi Ho
NeurIPS 2023, NeurIPS 2023 BUGS Workshop Oral
Paper

Proposed a backdoor attack on the diffusion models with broader settings, including score-based models, various samplers, and text triggers.

sym

How to Backdoor Diffusion Models?
Sheng-Yen Chou, Pin-Yu Chen, Tsung-Yi Ho
CVPR 2023, ICLR 2023 BANDS Workshop Best Paper Award
Paper

Proposed a new backdoor attack on the diffusion models. We found that the attacker can embed a backdoor into the diffusion models with 5% poison rate and 50 epoch fine-tuning.

sym

Elijah: Eliminating Backdoors Injected in Diffusion Models via Distribution Shift
Shengwei An, Sheng-Yen Chou, Kaiyuan Zhang, Qiuling Xu, Guanhong Tao, Guangyu Shen, Siyuan Cheng, Shiqing Ma, Pin-Yu Chen, Tsung-Yi Ho, Xiangyu Zhang
AAAI 2024, NeurIPS 2023 BUGS Workshop
Paper

Proposed backdoor detection and removal algorithms for backdoor diffusion models.

sym

Rethinking Backdoor Attacks on Dataset Distillation: A Kernel Method Perspective
Ming-Yu Chung, Sheng-Yen Chou, Chia-Mu Yu, Pin-Yu Chen, Sy-Yen Kuo, Tsung-Yi Ho
ICLR 2024
Paper

Proposed a computational-effiecient backdoor attack on data distillation based on NTK theory.

sym

Single-level Adversarial Data Synthesis based on Neural Tangent Kernels
Yu-Rong Zhang, Ruei-Yang Su, Sheng-Yen Chou, Shan-Hung Wu
ArXiv Preprint
Paper

Proposed a new generative model that reduce the bilevel optimization objective into single-level one while keeping the spirit of adversarial learning. Our method takes NTK-GP as the surrogate of the discriminator and generate comparable images with SOTA GANs.

  Projects
sym

EfficientDet
An EfficientDet implementation in TF2.0 based on the paper EfficientDet: Scalable and Efficient Object Detection on CVPR’20.

sym

DRL Collection
A collection of implements of classical DRL algorithms. It contain modular implementations of A3C, A2C, DDQN, and REINFORCE (naive) with Tensorflow2.0.

sym

ML Collection
Implemetation and derivation of ML algorithms, including SVM and VBGMM in Python.

sym

Implementation of 2V2PL
Implemented the 2V2PL concurrency protocol on VanillaDB with Java and improved the throughput at most 5 times than S2PL in TPCC workload

  Service
  Honors
  • 3rd place of 7th NTHU ENTREPRENEUR DAYS, Taiwan, 2019

  • Academic Excellence Award, NTHU, 2022