Fig. 3
From: MetaFlux: Meta-learning global carbon fluxes from sparse spatiotemporal observations
Schematic diagram of our (a) meta-learning approach that is meta-trained on data-abundant tasks to obtain a set of ϕ, and fine-tuned on data-sparse tasks to get θ*; (b) baseline algorithm that is not meta-trained. The optimal θ* in the latter case tends to be biased towards data-abundant tasks as represented by the gradient sets \({\phi }_{n}^{* }\).
