Fighter: Unveiling the Graph Convolutional Nature of Transformers in Time Series Modeling
This work provides a theoretical reinterpretation for researchers in time series analysis, offering clearer mechanistic insights, though it is incremental in building on existing Transformer and GCN concepts.
The paper tackled the opacity of Transformer mechanisms in time series modeling by establishing their equivalence to Graph Convolutional Networks, leading to the Fighter architecture that achieved competitive performance on forecasting benchmarks with improved interpretability.
Transformers have achieved remarkable success in time series modeling, yet their internal mechanisms remain opaque. This work demystifies the Transformer encoder by establishing its fundamental equivalence to a Graph Convolutional Network (GCN). We show that in the forward pass, the attention distribution matrix serves as a dynamic adjacency matrix, and its composition with subsequent transformations performs computations analogous to graph convolution. Moreover, we demonstrate that in the backward pass, the update dynamics of value and feed-forward projections mirror those of GCN parameters. Building on this unified theoretical reinterpretation, we propose \textbf{Fighter} (Flexible Graph Convolutional Transformer), a streamlined architecture that removes redundant linear projections and incorporates multi-hop graph aggregation. This perspective yields an explicit and interpretable representation of temporal dependencies across different scales, naturally expressed as graph edges. Experiments on standard forecasting benchmarks confirm that Fighter achieves competitive performance while providing clearer mechanistic interpretability of its predictions.