Currently, I am a senior software engineer in a stealth unicorn startup. Before that, I spent my unforgettable 4 years working with the incredibly talented advisor and her students in Secure Learning Lab at University of Illinois, Urbana-Champaign.
The lessons I learnt from them are and will always be guiding me through my career in the forseeable future.
Personal update: I am actively looking for a Phd advisor whose interest is aligned.
I am broadly interested in many topics in practition and learning theory. I have experience in computer vision, adversairal learning and desigining hardware friendly overparameterized models.
Some personal delusions:
- To scale the neural nets to the brain level, we need asynchronous local updates that not only work on papers but also in pratice. It's hard to imagine our brains update 86 billion neurons and 100 trillion connections, whenever we learn something new.
- To ground our models, we need fast adpative mechanisms that enable few-shot generalization. Just imagine Neurallink will have to be retrained on everyone's neurological data from scratch to perform simple tasks, such as moving a cursor.
Pixelated Butterfly: Simple and Efficient Sparse training for Neural Network Models. [In submission]
Beidi Chen, Tri Dao, Kaizhao Liang, Jiaming Yang, Zhao Song, Atri Rudra, Christopher Re
Uncovering the Connections Between Adversarial Transferability and Knowledge Transferability. [ICML 2021]
Kaizhao Liang*, Jacky. Y. Zhang*, Boxin Wang, Zhuolin Yang, Sanmi Koyejo, Bo Li
Adversarial Mutual Information for Text Generation [ICML 2020]
Boyuan Pan, Yazheng Yang, Kaizhao Liang, Bhavya Kailkhura, Zhongming Jin, Xian-Sheng Hua, Deng Cai, Bo Li
Unrestricted Adversarial Examples via Semantic Manipulation [ICLR 2020]
Anand Bhattad*, Min Jin Chong*, Kaizhao Liang, Bo Li, David A. Forsyth
Gesture Control of autonomous vehicle
a new interface of human interaction with Autonomous Vehicle.
Interval bound propagation pytorch
Exploring to train certified defense against adversarial attacks with limited resources by "tighter" objective
Learning to Run
TRPO and DDPG implementation to teach robot to walk in simulations.
CVPR2021, ICCV2021, ICLR2022, CVPR2022