A programmer for the fun of it.
I am now a PhD student in The Chinese University of Hong Kong, Shenzhen (CUHKSZ) (opens new window), advised by Prof.Tianshu Yu (opens new window). Before that, I received my BEng and MEng in Computer Science from the Southern University of Science and Technology (SUSTech) (opens new window) and Harbin Institute of Technology (Joint Master Program with SUSTech) (opens new window), advised by Prof.Qi Wang (opens new window).
- Machine Learning for Boolean Satisfiability Problem (SAT)
- Industrial SAT instance generation
- Learning-aided SAT solver
- Algorithm selection/configuration
W2SAT: Learning to generate SAT instances from Weighted Literal Incidence Graphs
In this paper, we propose W2SAT, a framework to generate SAT formulas by learning intrinsic structures and properties from given real-world/industrial instances implicitly. We introduce a novel SAT representation called Weighted Literal Incidence Graph (WLIG), which exhibits strong representation ability and generalizability against existing counterparts, and can be efficiently generated via a specialized learning-based graph generative model. Decoding from WLIGs into SAT problems is then modeled as finding overlapping cliques with a novel hill-climbing optimization method termed Optimal Weight Coverage (OWC).
SAT-LLL: Lovász Local Lemma for SAT
In this paper, we introduce SAT-LLL, a tool that applies LLL to the Boolean Satisfiability (SAT) problem. The SAT-LLL consists of an instance generator that produces formulas under the LLL condition (in the local lemma regime), an algorithm that decides if a given SAT instance is in the local lemma regime, and a solver that can search for solutions or uniformly sample solutions in the local lemma regime. Overall, SAT-LLL is a valuable tool bridging the Lovász Local Lemma and the SAT problem by simplifying the simulation and experimentation of future research. Being a valuable tool for researchers and practitioners, SAT-LLL offers a novel approach to studying SAT problems from an LLL perspective.
1. A steganography scheme based on adversarial examples (master thesis)
In this work, we propose a friendly application constructed by adversarial examples: a steganography scheme based on adversarial examples. In this scheme, we regard the adversarial nature of the adversarial examples itself as information, which can be decoded by the specific model only.
2. Multi-task Learning for Aspect-based Sentiment Analysis (bachelor thesis)
This work proposes a deep multi-task learning framework for aspect term extraction (ATE). Furthermore, it presents a novel way to combines aspect term extraction and aspect sentiment classification into an individual task, which can avoid the error propagation from the pipeline approach.
ServingAgent is designed as a middleware for model serving between web server and model server to help the server improve the GPU utilization then speedup online inference.
ServingTemplate is a tool to auto gereate the model serving project template.
Awards & Honors
- University Scholarship, SUSTech 2018 – 2019
- College Student Start-up Scholarship, SUSTech 2014 – 2017
- Tool: Git, Docker, Kubernetes, etc.