FastKAN
An efficient implementation of Kolmogorov-Arnold Networks that replaces the fixed activation functions of MLPs with learnable univariate functions parameterized by B-splines. It is well suited to fitting smooth continuous mappings and to continuous control signal processing.
Timeline (last 90 days)
PCEI is an open-source project whose full name is Physically Constrained Emergent Intelligence, combining three technologies—BitNet 1.58b, Mamba and FastKAN—for on-device embodied intelligence.
FastKAN is an efficient implementation of the Kolmogorov-Arnold network that uses B-splines to replace the fixed activation functions of traditional MLPs, and in PCEI it is used for continuous motion control in the MuJoCo physics engine while handling 4 control streams.
PCEI's experiments are currently concentrated in the MuJoCo simulation environment, and whether the spline-fitting capability FastKAN demonstrates in simulation can generalize to the noisy environment of real robots remains unknown.
All Facts (3)
PCEI is an open-source project whose full name is Physically Constrained Emergent Intelligence, combining three technologies—BitNet 1.58b, Mamba and FastKAN—for on-device embodied intelligence.
50%UnverifiedFastKAN is an efficient implementation of the Kolmogorov-Arnold network that uses B-splines to replace the fixed activation functions of traditional MLPs, and in PCEI it is used for continuous motion control in the MuJoCo physics engine while handling 4 control streams.
50%UnverifiedPCEI's experiments are currently concentrated in the MuJoCo simulation environment, and whether the spline-fitting capability FastKAN demonstrates in simulation can generalize to the noisy environment of real robots remains unknown.
50%