2
$\begingroup$

I have a data set and few human rules. I want to learn the machine learning system. One of the paper I came across with is:

Harnessing Neural Network

Above paper trained some teacher network to learn the system in order to improve the accuracy but they still require enough data. But my requirement is different. It may be possible that we have very few data points corresponding to a rule. How can I design a system which can utilize the both learning from data set and human rules.

$\endgroup$
1
  • 1
    $\begingroup$ Encode your knowledge into Bayesian priors. If you need more speciific advice, explain the nature of your problem and 'human rules'. $\endgroup$
    – Emre
    Dec 1 '17 at 8:03
1
$\begingroup$

The area of combining machine learning and human knowledge is called expert-augmented machine learning.

One common method is called RuleFit. RuleFit has two steps:

  1. Fit a tree-based model to make rules.
  2. Fits a linear model with these rules as input.

The result is highly interpretable which can be filtered or applied by humans.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.