Reward, punishment and internal expectation for training the random neural network with reinforcement

Halici U.

13th International Symposium on Computer and Information Sciences (ISCIS 98), BELEK ANTALYA, Türkiye, 26 - 28 Ekim 1998, cilt.53, ss.162-169 identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 53
  • Basıldığı Şehir: BELEK ANTALYA
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.162-169
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

The reinforcement learning scheme proposed in [8] for random neural networks [5] is based on reward and performs well for a stationary environment. However, when the environment is not stationary extinction becomes an important problem to be considered. In this paper, the reinforcement learning scheme is extended by introducing a weight update rule that takes into consideration the internal expectation of reinforcement. Such a scheme has made extinction possible while resulting in a good convergence to the most rewarding action.