A Generic Graph Sparsification Framework using Deep Reinforcement Learning
This addresses the resource-intensive processing of complex graphs for applications in data analysis and network optimization, offering a flexible and efficient solution, though it builds incrementally on reinforcement learning techniques.
The paper tackles the problem of graph sparsification, where edge-reduced graphs are produced while preserving user-defined metrics, and presents SparRL, a deep reinforcement learning framework that outperforms existing methods in producing high-quality sparsified graphs across various objectives.
The interconnectedness and interdependence of modern graphs are growing ever more complex, causing enormous resources for processing, storage, communication, and decision-making of these graphs. In this work, we focus on the task graph sparsification: an edge-reduced graph of a similar structure to the original graph is produced while various user-defined graph metrics are largely preserved. Existing graph sparsification methods are mostly sampling-based, which introduce high computation complexity in general and lack of flexibility for a different reduction objective. We present SparRL, the first generic and effective graph sparsification framework enabled by deep reinforcement learning. SparRL can easily adapt to different reduction goals and promise graph-size-independent complexity. Extensive experiments show that SparRL outperforms all prevailing sparsification methods in producing high-quality sparsified graphs concerning a variety of objectives.