Research

Subgraph mining

2021.09 – Present, National Supercomputing Center In Changsha, Changsha, Hunan, China

1. Maximum K-Biplex Search. This research proposes a new dense subgraph model (x,y)-core and an improved core-based search framework. Sequentially, this research applies a symmetric branch-and-bound search algorithm. Additionally, an FPGA-based heuristic algorithm is proposed.
2. Structural Hypergraph Clustering. This research proposes a novel hyperedge-centric structural clustering model with improved clustering quality. To efficiently cluster hypergraphs, this research proposes two indexes: the order-index and the lightweight similarity-bucket index. In addition, a sequential algorithm and a parallel algorithm are proposed to query from the indexes.
3. Structural Graph Clustering. Given that existing state-of-the-art indexes fail to fully leverage stored information and accelerate all phases of SCAN, this research reorganizes stored data based on SCAN’s characteristics and investigates two indexes requiring reduced space overhead and query time.
4. Distributed K-Truss Decomposition. This research proposes a distributed k-truss decomposition algorithm based on a distributed graph processing system, Quegel.

AI4DB

2024.07 – 2025.10, OceanBase, Ant Group

1. Query Plan Cost Estimation for OceanBase. This research designs corresponding encoding schemes and a two-layer Transformer model for complex filters commonly encountered in OceanBase, ultimately achieving an index-aware AI Cost estimator tailored for complex filtering tasks.
2. RL-Based Index Selection. This research applies the PPO2 model and the SAC model to implement index selection.