Search papers, labs, and topics across Lattice.
This paper introduces GraphPL, a novel approach to unsupervised modality imputation in patchwork learning settings where clients have access to varying subsets of modalities. GraphPL leverages graph neural networks to effectively integrate information from all observed modalities, addressing the limitations of existing methods that rely on only a subset. Experiments on benchmark and real-world EHR datasets demonstrate that GraphPL achieves state-of-the-art imputation performance and learns strong downstream features for tasks like disease prediction.
Patchwork learning gets a boost: GraphPL uses GNNs to flexibly integrate all observed modalities, achieving SOTA imputation performance even with noisy inputs.
Current research on distributed multi-modal learning typically assumes that clients can access complete information across all modalities, which may not hold in practice. In this paper, we explore patchwork learning, in which the modalities available to different clients vary, and the objective is to impute the missing modalities for each client in an unsupervised manner. Existing methods are shown not to fully utilize the modality information as they tend to rely on only a subset of the observed modalities. To address this issue, we propose GraphPL, which combines graph neural networks with patchwork learning to flexibly integrate all observed modalities and remains robust with noisy inputs. Experimental results show that GraphPL achieves SOTA performance on benchmark datasets. Our results on real-world distributed electronic health record dataset show GraphPL learns strong downstream features and enables tasks like disease prediction via superior modality imputation.
