1: \begin{abstract}

2: In this paper, we introduce a new machine learning theory

3: based on multi-channel parallel adaptation for rule discovery. 

4: This theory is distinguished from the familiar parallel-distributed

5: adaptation theory of neural networks in terms of channel-based

6: convergence to the target rules.

7: We show how to realize this theory in a learning system named CFRule. 

8: CFRule is a parallel weight-based model, but it

9: departs from traditional neural computing

10: in that its internal knowledge is comprehensible. Furthermore,

11: when the model converges upon training,

12: each channel converges to a target rule.

13: The model adaptation rule is derived by multi-level parallel weight

14: optimization based on gradient descent.

15: Since, however, gradient descent only guarantees local optimization, 

16: a multi-channel regression-based optimization strategy is developed

17: to effectively deal with this problem.

18: Formally, we prove that the CFRule model can explicitly and precisely

19: encode any given rule set. Also, we prove a property related to

20: asynchronous parallel convergence, which is a critical element

21: of the multi-channel parallel adaptation theory for rule learning.

22: Thanks to the quantizability nature of the CFRule model, rules

23: can be extracted completely and soundly via a threshold-based mechanism.

24: Finally, the practical application of the theory is demonstrated

25: in DNA promoter recognition and hepatitis prognosis prediction.

26: \end{abstract}

27: