SHyPar: A Spectral Coarsening Approach to Hypergraph Partitioning

Hamed Sajadinia,Ali Aghdaei,Zhuo Feng

2024-10-09

Abstract:State-of-the-art hypergraph partitioners utilize a multilevel paradigm to construct progressively coarser hypergraphs across multiple layers, guiding cut refinements at each level of the hierarchy. Traditionally, these partitioners employ heuristic methods for coarsening and do not consider the structural features of hypergraphs. In this work, we introduce a multilevel spectral framework, SHyPar, for partitioning large-scale hypergraphs by leveraging hyperedge effective resistances and flow-based community detection techniques. Inspired by the latest theoretical spectral clustering frameworks, such as HyperEF and HyperSF, SHyPar aims to decompose large hypergraphs into multiple subgraphs with few inter-partition hyperedges (cut size). A key component of SHyPar is a flow-based local clustering scheme for hypergraph coarsening, which incorporates a max-flow-based algorithm to produce clusters with substantially improved conductance. Additionally, SHyPar utilizes an effective resistance-based rating function for merging nodes that are strongly connected (coupled). Compared with existing state-of-the-art hypergraph partitioning methods, our extensive experimental results on real-world VLSI designs demonstrate that SHyPar can more effectively partition hypergraphs, achieving state-of-the-art solution quality.

Social and Information Networks,Machine Learning

What problem does this paper attempt to address?

### What problems does this paper attempt to solve?

This paper aims to solve the key challenges in **Hypergraph Partitioning**. Specifically, existing hypergraph partitioning methods usually rely on simple heuristic algorithms for coarsening, and these methods often overlook the structural characteristics of hypergraphs, resulting in poor partitioning effects.

#### Main problems:
1. **Limitations of existing methods**: Traditional hypergraph partitioners use a multi - layer paradigm to construct successively coarser hypergraphs and guide cut optimization at each layer. However, most of these methods are based on simple heuristic algorithms of local structures (such as vertex similarity or hyper - edge similarity) and fail to fully consider the global structural characteristics of hypergraphs.

2. **Optimizing the quality of hypergraph partitioning**: In order to improve the quality of hypergraph partitioning, a new framework is needed that can better capture the global structural information of hypergraphs, thereby achieving better partitioning results.

#### Solutions:
The paper introduces a multi - layer spectral framework named **SHyPar** to partition large - scale hypergraphs by using the effective resistance of hyper - edges and flow - based community detection techniques. The main innovations of this framework include:

- **Spectral Coarsening**: By combining the effective resistance of hyper - edges and a flow - based local clustering scheme, SHyPar can generate a smaller hypergraph representation while maintaining the key structural properties of the hypergraph.

- **Improved partitioning quality**: Compared with the existing state - of - the - art hypergraph partitioning methods, SHyPar shows higher partitioning quality and efficiency on actual VLSI design data sets.

#### Specific goals:
- **Reduce the number of hyper - edges across partitions (cut size)**: Improve the partitioning effect by minimizing the number of hyper - edges between different partitions.

- **Maintain balance constraints**: Ensure that the difference in node or hyper - edge weights in each partition does not exceed a given threshold to meet the requirements in practical applications.

In summary, this paper is committed to developing a new hypergraph partitioning framework based on spectral theory to overcome the limitations of existing methods and significantly improve the quality and efficiency of hypergraph partitioning.