作者: G. P. Drago , M. Morando , S. Ridella
DOI: 10.1007/BF01414646
关键词:
摘要: An algorithm for fast minimum search is proposed, which achieves very satisfying performance harmonising the Vogl's and Conjugate Gradient algorithms. Such effectiveness achieved by making adaptive, in a simple satisfactory way, both learning rate momentum term, executing controls corrections on possible cost function increase moves opposite to direction of negative gradient. Thanks these improvements, we can obtain good scaling relationship learning. As regards real world context, musical application showed favourable results: besides convergence speed, high generalisation capability has been achieved, as confirmed subjective evaluations objective tests.