论文部分内容阅读
提出一种二次不动—惩罚变结构随机自动机模型（Q(IP)）。较之于其线性形式（L(IP)），新模型的学习带有一定的自信（当然，有时也可能是自负）。特别，跟传统自动机不同的是，新算法的极限行为同时兼具吸收壁和遍历性。
A quadratic immutability-punishing structure random automaton model (Q (IP)) is proposed. Compared to its linear form (L (IP)), the new model has some level of confidence in learning (of course, sometimes it may be conceited). In particular, unlike traditional automata, the limit behavior of the new algorithm combines both absorption walls and ergodicity.