Skip to main content
. Author manuscript; available in PMC: 2023 Feb 1.
Published in final edited form as: IEEE Trans Knowl Data Eng. 2020 Apr 22;34(2):531–543. doi: 10.1109/tkde.2020.2989405

Fig. 1:

Fig. 1:

CHEER: (a) a rich model was first built using rich multimodal or multi-channel data; (b) the behaviors of rich model are then infused into the poor model using paired data (i.e., behavior infusion); and (c) the poor model is trained to fit both the rich model’s predictions on paired data and its poor dataset (i.e., target infusion).