The Evolution Of Variable Learning Rates

Neural plasticity in humans is well known to be age dependent, with 'critical periods' for the learning of many tasks. It is reasonable to hypothesise that this has some intrinsic advantage over constant plasticity, and that it has arisen as the result of evolution by natural selection. If this is true, then it may also prove useful for building more efficient artificial systems that are required to learn how to perform appropriately. In this paper I explore these ideas with a series of explicit evolutionary simulations of some simplified control systems.

[1]  J. Baldwin.  A New Factor in Evolution , 1896, The American Naturalist.

[2]  Robert A. Jacobs,et al.  Increased rates of convergence through learning rate adaptation , 1987, Neural Networks.

[3]  C. Schor,et al.  Negative feedback control model of proximal convergence and accommodation , 1992, Ophthalmic & physiological optics : the journal of the British College of Ophthalmic Opticians.

[4]  B. Julesz,et al.  Maturational Windows And Adult Cortical Plasticity , 1995 .

[5]  Xin Yao,et al.  Evolving artificial neural networks , 1999, Proc. IEEE.

[6]  John A. Bullinaria.  Exploring the Baldwin Effect in Evolving Adaptable Control Systems , 2000, NCPW.

[7]  John A. Bullinaria,et al.  Neural Network Control Systems that Learn to Perform Appropriately , 2001, Int. J. Neural Syst..