Zihao Wang

mypic.jpeg
University Center 101, HKUST, Clear Water Bay

Bibliography

I am Zihao Wang （王子豪）, a Ph.D. candidate from CSE department, HKUST since Sept. 2020. My advisor is Prof. Yangqiu Song. I obtained my master’s degree in Computer Science and Technology in 2020 with Prof. Yong Zhang, my primary bachelor’s degree in Energy and Power Engineering in 2017, and my secondary bachelor’s degree in Pure and Applied Mathematics in 2018 with Prof. Hao Wu, all from Tsinghua University.

From Aug. 2023 to Jan. 2024, I visited the IDEA lab @ UIUC under the supervision of Prof. Hanghang Tong.



Research interests

The general theme of my research is reasoning with data, with special interests in logic on graphs, language models, optimal transport, and AI for sciences.

Summary at the end of 2023

(click to expand)
#### First-order reasoning on KG During my PhD, I studied graph problems justified by the first-order language. $$(\text{Reason}, \text{Data}) = (\text{First-order language}, \text{Graph})$$ My work includes: - On the language aspect, I build empirical datasets and benchmarks to cover the Existential First Order (EFO) [NeurIPS'21, arXiv:2304.07063, arXiv:2307.13701] - On the model aspect, I build deep learning and representation learning methods to answer first-order queries [NAACL'22, ACL'23, ICLR'23] #### Natural language reasoning (beyond the first-order language) My research work also involves reasoning with natural languages, which is beyond the first order. $$(\text{Reason}, \text{Data}) = (\text{Natural language}, \text{World knowledge})$$ My research can be roughly separated by knowledge representation and reasoning - World knowledge representation: Commonsense knowledge representation [EMNLP'22a] - World knowledge reasoning: Neurosymbolic reasoning for natural language query [EMNLP'22b] ✨ My recent interest includes the reasoning capability of multi-agent LLMs. (arXiv:2311.07076) #### Data modeling (within the first-order language) I frame the machine learning tasks as a single predicate $$ \text{Prediction}(\text{Input}, \text{Output}) $$ $$(\text{Reason}, \text{Data}) = (\text{Single Predicate}, \text{Multimodal data})$$
Shallow reasoning by data modeling ✨ My recent interest includes on modeling of molecule graphs. (arXiv:2310.03152)
The predicate IsLabelOf (supervised learning)
  • Understanding deep learning tricks [arXiv:2010.12648]
  • Designing neural heuristics for applications [AJODO'23, JORC'23]
  • Compressing deep neural networks [ICML'23]
The predicate IsLatentCodeOf (unsupervised learning)
  • Understanding the variational auto encoder [NeurIPS'22]
  • Modeling temporal encoding/decoding process [VTC'21]
  • ML for hardware design [ICCT'21]
The predicate IsSameDistribution (optimal transport)
  • Efficient algorithms [JSC'23,CSIAM-AM'23,CMS'22]
  • Application to point cloud alignment [ACL'20, EMNLP'20, COLING'22]
  • Application to cross-domain recommendation [CIKM'22]



Academic services

  • Conference Reviewer:
    • NLP: EMNLP 2021 - 2023; ACL 2023 - 2024; EACL 2024, NAACL 2024;
    • ML: ICLR 2024; ICML 2024; NeurIPS 2023; NeurIPS dataset & benchmark 2023.
    • Others: AAAI 2024; CVPR 2023; KDD 2023 - 2024; CIKM 2023;
  • Journal Reviewer: TNNLS, TASLP, TIST, Information Fusion, DMLR