ConceptFisher Information Matrix / FIM
Fisher Information Matrix
A matrix in information geometry that measures the curvature of the parameter space of a family of probability distributions. Under a Riemannian manifold framework, it acts as a metric tensor defining distances between distributions and serves as the central mathematical tool in natural gradient descent.
Timeline (last 90 days)
Oct 9
EWC was proposed by Kirkpatrick et al. in 2017; its theoretical foundation comes from Bayesian inference, using the Fisher information matrix to estimate the importance of each weight to old tasks.
Unverified50%