authoredinburghinformaticsepsrc-fellowspectral-clusteringdirected-graphsparsification

He Sun

Affiliation: School of Informatics, University of Edinburgh.

Funding: EPSRC Early Career Fellowship (EP/T00729X/1).

Email: h.sun@ed.ac.uk

Contributions to this wiki

  • Senior author of entities/Laenen Sun 2020 Higher-Order Spectral Clustering (NeurIPS 2020) — developed the Hermitian Laplacian theoretical framework for directed-graph spectral clustering, with Cheeger-style approximation guarantees.
  • Co-author of entities/Cucuringu Li Sun Zanetti 2020 Hermitian Adjacency (AISTATS 2020) — the Hermitian adjacency i(A - A^\top) approach (predecessor to Laenen-Sun 2020); same theoretical lineage but different mathematical construction (no phase factor, \lceil k/2 \rceil eigenvectors).
  • Co-author of Li, Sun, Zanetti (ESA 2019) "Hermitian Laplacians and a Cheeger inequality for the Max-2-Lin problem" — earliest Cheeger-style result linking Hermitian spectral theory to combinatorial optimization.
  • Co-author of Peng, Sun, Zanetti (SIAM J. Comput. 2017) — the foundational spectral clustering paper "Partitioning well-clustered graphs: Spectral clustering works!" that Laenen-Sun 2020 extends.
  • Co-author of Sun & Zanetti (ACM TOPC 2019) "Distributed graph clustering and sparsification" — spectral sparsification techniques used in SimpleHerm.

Research themes

  • Spectral graph theory (directed / undirected)
  • Graph clustering and sparsification
  • Approximation algorithms for combinatorial optimization via spectral methods
  • Distributed computing on graphs

Importance to this wiki

He Sun is the central methodological link between:

  1. Undirected spectral clustering theory (Peng-Sun-Zanetti 2017) →
  2. Hermitian adjacency for directed graphs (Cucuringu-Li-Sun-Zanetti 2020) →
  3. Hermitian Laplacian with chain structure (Laenen-Sun 2020) →
  4. Financial lead-lag applications (Bennett-Cucuringu-Reinert 2022, citing Laenen-Sun as alternative)

Any future work on directed-graph spectral clustering for financial networks should engage with his lab's work.

See also