Conge 精进

KBAI 笔记 10 Incremental Concept Learning

本文 1907 字,阅读全文约需 6 分钟



Exercise: Identifying a Foo I


  • background knowledge is important to make the judgement whether the fourth graph shows the concept of foo.

Exercise: Identifying a Foo II

quiz 2

Exercise: Identifying a Foo III

quiz 2

  • yes or no answer could be correct. yes is a generalization and no is a specification
  • learning is incremental (learn from one example at a time)
  • examples are labeled.
  • examples come from specific order
  • differ from K space reasoning
  • the number of examples is small.
  • what to learn is going to be a hard question( over generalization or over specification)

Incremental Concept Learning

Process of Incermental concept learning

Example of over generalization and over specification using children as example. what is a dog.


AI representation of an Arch.

Generalization to Ignore Features

New Arch example

Drop-link heuristic

Specialization to Require Features

require-link heuristic

Specialization to Exclude Features

"forbid-link" heuristic

Generalization to Abstract Features

"enlarge" heuristic

Generalization with Background Knowledge

"clime tree" heuristic

An Alternative Visualization

Heuristics for Concept Learning

Summary of Heuristics

Exercise: Re-Identifying a Foo I

Exercise: Re-Identifying Foo II

Exercise: Re-Identifying Foo III

The concept of block can be brick or Cylinder.

backgroup knowledge

Exercise: Re-Identifying Foo VI

Do nothing

  • the example is not a concept and the current COncept is already exclusive. we don’t need to do anything or learn anything.

Final Concept of a Foo

  • when there is limited number of examples, Agent has to use its background knowledge and use all the examples to incrementally learn the concept.


Wrap Up

The Cognitive Connection

In real life, cognitive agent like humans are usually given one example after another. Incremental concept learning is the closed way of how human learn new concept.