当前位置: X-MOL 学术Complex Intell. Syst. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
A kind of artificial intelligence model based on nature without statistic
Complex & Intelligent Systems ( IF 5.0 ) Pub Date : 2020-07-09 , DOI: 10.1007/s40747-020-00174-z
Jun Meng , Jie-zhou Yu , Xiao Chen

Simple rules can generate complexity and human can also learn fast with rules. Can machines learn in a similar way? Can artificial intelligence be independent of statistics? Machine learning is growing rapidly but models are poorly interpretable and depend on statistics. We propose a method by iteration based on causality which is the real one exists in the system. It is composed of fixed goals and basic rules called DNA rules. These DNA rules can be obtained from the definition and is not statistical rules. The causality in rules promises the process to be precise, because the potential attractor of the system is deterministic, because it is subjected to the rules although the system is complex especially under uncertain interference. Such a model not only works well in the traditional deterministic systems like the stable point and limited circle but also can work in some seemingly random and systems which are considered to be stochastic systems. The model is taken to play a game and it makes the machine learns fast and adaptively, and it is also interpretable with the causality and independent from the amount of data for it is based on causal iteration. It learns and even predicts the seemingly random interference in the game. We found such a model is adaptable, and it works well even in out-of-sample situations. The model is compared with an LSTM network in prediction a seemingly random sequence, the result shows the causality-based model also works well. We think that it may help to solve some problems hard for the traditional statistical method and become an enrichment for the current models.



中文翻译:

一种基于自然的无统计人工智能模型

简单的规则会产生复杂性,人类也可以通过规则快速学习。机器可以类似的方式学习吗?人工智能可以独立于统计数据吗?机器学习正在迅速发展,但是模型难以解释并且依赖于统计数据。我们提出了一种基于因果关系的迭代方法,该因果关系是系统中真正存在的一种因果关系。它由固定的目标和称为DNA规则的基本规则组成。这些DNA规则可以从定义中获得,而不是统计规则。规则中的因果关系保证了过程的精确性,因为系统的潜在吸引者是确定性的,因为尽管系统复杂,尤其是在不确定的干扰下,但它仍受规则约束。这样的模型不仅可以在诸如稳定点和有限圆之类的传统确定性系统中很好地运行,而且可以在某些看似随机的系统中被认为是随机系统。该模型是用来玩游戏的,它使机器能够快速,自适应地学习,并且还可以使用因果关系解释该因果关系,并且它基于因果迭代而不受数据量的影响。它学习甚至预测游戏中看似随机的干扰。我们发现这样的模型是适应性强的,即使在样本不足的情况下它也能很好地工作。该模型与LSTM网络进行了比较,预测了一个看似随机的序列,结果表明基于因果关系的模型也能很好地工作。

更新日期:2020-07-09
down
wechat
bug