How PEIL learns unobservable parameters from signal recovery (IMAGE)
Caption
PEIL addresses inverse problems in which recovering a target signal from indirect, noisy observations depends on unobservable physical parameters. A neural network predicts the parameters required by a fixed, differentiable physical solver, which then recovers the target signal. The recovery error is propagated back through the solver to train the parameter-estimation network without parameter labels.
Credit
©Science China Press
Usage Restrictions
Use with credit.
License
Original content