Search papers, labs, and topics across Lattice.
Tianjin University
1
0
0
Learning a residual correction to a physics-informed estimate dramatically boosts QNN performance in cloud-cover parameterization, outperforming both direct QNNs and compact MLPs.