GLAudio Listens to the Sound of the Graph
This introduces a new paradigm for graph learning that could benefit researchers in graph neural networks by addressing fundamental bottlenecks like over-smoothing and over-squashing.
The paper tackles the problem of learning on graph-structured data by proposing GLAudio, a novel architecture that separates information propagation and processing into distinct steps using audio wave signals, achieving theoretical characterization of expressivity and experimental investigation of over-smoothing and over-squashing on various datasets.
We propose GLAudio: Graph Learning on Audio representation of the node features and the connectivity structure. This novel architecture propagates the node features through the graph network according to the discrete wave equation and then employs a sequence learning architecture to learn the target node function from the audio wave signal. This leads to a new paradigm of learning on graph-structured data, in which information propagation and information processing are separated into two distinct steps. We theoretically characterize the expressivity of our model, introducing the notion of the receptive field of a vertex, and investigate our model's susceptibility to over-smoothing and over-squashing both theoretically as well as experimentally on various graph datasets.