Sample-complexity bound for high-order linear systems
Produced by Long-Context Linear System Identification
Learning without mixing
A statistical phenomenon in which learning from dependent observations avoids a multiplicative sample-complexity penalty from the mixing time.
A statistical phenomenon in which learning from dependent observations avoids a multiplicative sample-complexity penalty from the mixing time.
Learning without mixing does not assert that the observations are independent or that the dynamics mix rapidly. It means that the learning rate retains essentially the full effective sample size instead of being reduced by a mixing-time factor.