87f476af4053961667c2c08e9f4b850e-Supplemental-Conference.pdf

We performed all experiments on 4 V100 GPUs. Apart from the root mean square error (RMSE), we also report the energy spectrum error (ESE) forocean current prediction which quantifies the physical consistency. Our model achieves this by learning on multiple tasks simultaneously and then adapting to new tasks with domain transfer. As these are upper bounds, they do not necessarily imply better generalization, although empirically, this is what we observe. Suppose we haveK tasks{ck}Kk=1 C, each of which is sampled from acontinuous parameter spaceC.