Hanzhi Wang   /   王涵之

I am a PostDoc at BARC (Basic Algorithms Research Copenhagen), University of Copenhagen, working with Prof. Mikkel Thorup. I obtained my Ph.D. in Computer Science and Technology from the School of Information at Renmin University of China in May 2024, under the supervision of Prof. Zhewei Wei. I received my B.E. degree in Computer Science and Technology from the School of Information at Renmin University of China in June 2019.

hanzhi.hzwang@gmail.com  /  Github  /  Google Scholar

My research interests are graph analysis algorithms, with a particular focus on the design of scalable algorithms for large graphs. My previous work primarily centers on sublinear-time algorithms for random-walk probability estimation. I am also interested in local clustering and graph neural networks.

Note: Papers marked with ** use alphabetic ordering of authors, following the convention of theoretical computer science.

This paper improves the time dependence on L_max in solving the problem of estimating effective resistance (ER) on undirected graphs. Here, L_max is a graph parameter influenced by the mixing time of random walks on the graph. We emphasize that the value of L_max in real-world networks can be very large. However, previous work often assumes a small value for L_max, which can result in significant approximation errors. We hope this paper draws attention to the critical role of L_max in ER approximation.

We study the fundamental subset sampling problem in a general dynamic setting and achieve optimal complexity results for both query and update time simultaneously.

We establish matching upper and lower bounds for the worst-case computational complexity of single-node PageRank computation on undirected graphs, using surprisingly simple algorithms and analysis.

We recap several frequently used techniques for Personalized PageRank (PPR) computation and conduct a comprehensive survey of various recent PPR algorithms from an algorithmic perspective.

We establish matching upper and lower bounds for the worst-case computational complexity of single-node PageRank computation on directed graphs. Additionally, for the celebrated ApproxContributions algorithm (also known as Backward Push or Local Push, proposed by Andersen, Borgs, Chayes, Hopcroft, Mirrokni, and Teng in WAW 2007), we upper bound its worst-case computational complexity on directed graphs and prove its optimality by establishing a matching lower bound.

For the problem of estimating a single node's PageRank score in an undirected graph, we tighten the upper bound for query time complexity from O((nd_t)^1/2) to O(min(d_t, m^1/2)) (omitting logarithmic factors). Here n and m denote the number of nodes and edges in the given graph, respectively. d_t denotes the degree of the given target node t.

We study the fundamental problem of sampling indepedent events (called subset sampling) in a dynamic setting. We present the ODSS algorithm, which is the first to achieve the optimal time complexity per sample and per update simultaneously.

The state-of-the-art algorithm, LocalPush, for Personalized PageRank computation can be rather inefficient on weighted graphs. In this paper, we propose an Edge-based Push Method (EdgePush), which decomposes the push operation of LocalPush into separate edge-based push operations and achieves superior query efficiency over LocalPush on weighted graphs.

We propose Instant Graph Neural Network (InstantGNN), an incremental computation approach to support instant updates of graph representation results on dynamic graphs.

We propose Approximate Graph Propagation (AGP), a unified randomized algorithm that computes various proximity queries and GNN feature propagation in almost optimal time complexity.

This work is an extended version of the conference paper "Exact Single-Source SimRank Computation on Large Graphs" accepted by SIGMOD 2020. With the ground truths computed by ExactSim, we conduct the first experimental study of the accuracy/cost trade-offs of existing approximate SimRank algorithms on large graphs.

We propose Randomized Backward Search (RBS), a novel algorithm that answers approximate single-target personalized PageRank queries (also known as PageRank contribution queries) with near optimal time complexity.

We propose ExactSim, the first algorithm that enables probabilistic exact single-source SimRank queries on large graphs. ExactSim can provide the ground truth with a precision up to 7 decimal places for single-source SimRank queries on large graphs within a reasonable query time.

PC member/Reviewer:

Last update: Dec. 2024      Template