β Scribed by Stefan Wimbauer; Nikolaus Klemmer; J. Leo van Hemmen
No coin nor oath required. For personal study only.
Unlearmng, a reverse process to learnmg accordmg to Hebb's rule, is a local and unsupervtsed procedure that gtves rtse to a substanttal improvement of the retrwval properties of an assoctative neural network (i) an enhancement of both the storage capactty and the domains of attractton, (li) the posstbthty to store correlated patterns, and ( lii ) the capabihty to dlstmgutsh between patterns and nonretrteval states Three dtfferent versions of this type of algorithm are introduced and the common underlymg mechamsms are explained Furthermore, unlearnmg ~s apphed to the storage of temporal sequences of correlated patterns that have been learned m a purely Hebblan way
π SIMILAR VOLUMES
The process of unlearning can serve as a basis for removing old knowledge structures. Unlearning makes it possible for new knowledge to be accepted, and for old structures to be changed or removed. However, since organizations have difficulties in changing when in fact they are successful, the trans