[KongchangAI]
ConceptSigmoid Reparameterization

Sigmoid Reparameterization

A parameterization technique that uses the Sigmoid function to map unconstrained real-valued parameters into the (0,1) interval, commonly used in continuous relaxation and gradient backpropagation.

Timeline (last 90 days)

Oct 17

Sigmoid reparameterization maps unconstrained real-valued parameters to the (0,1) interval, naturally satisfying the boundary constraints after relaxation and enabling gradients to propagate back to the parameters through the chain rule.

Unverified50%

All Facts (1)

Source Articles