Sarah Moore.pdf - KBAI Assignment 1 Addressing Ravens...

This preview shows page 1 - 3 out of 5 pages.

KBAI: Assignment 1: Addressing Raven’s Progressive Matrices Sarah Elizabeth Moore (smoore8) The basic problem to be solved relates to development of a knowledge-based AI agent to succeed in the Raven’s Progressive Matrices (RPM) test, which is an intelligence test based on visual analogy. In the lectures, RPM was broken down into at least three basic forms: 2x1, 2x2, and 3x3 matrices of visual analogies. The most basic set of progressive matrices, 2x1, is linear; the visual analogy to be formed is that of A:B::C:D. However, in the project objective, the KBAI agent must solve for 2x2 RPM, which is a square. It not only implies the same correspondence of 2x1 in that A:B::C:D, but it must also satisfy the correspondence of A:C::B:D. The analogous results may be similar to that of 2x1, but rely on extra steps during analysis and testing, especially from the more complicated issues. In class, the agent described dealt with 2x1 matrices. With 2x2 matrices, it becomes more difficult to figure out the solution because you have to satisfy not just one but two analogies. This means, the nodes and links may be harder to represent because there are two correspondences. The selection of a solution may also be harder due to the rules of the two analogies ruling out many potential candidates. Also, one would consider whether one of the analogies should have a greater weight when determining the correct candidate to complete the matrix. Still further, the semantic network rules that would apply to the score (e.g., reflection, rotation, scale, etc.) may be harder to determine when there is both a vertical and a horizontal spatial analogy.
The cognitive system to be designed will have to both be able to represent the visuals with verbal and visual representations, be able to generate the rules associated with the visual analogies, and be able to test candidates to complete the analogy. To create the agent, the following will be necessary: The basic steps of the agent will include: 1. Parsing the visual question (image-based or verbal-based) for base visuals 2. If image-based, computation of elements/objects in the image 3. Creation and storage of nodes that represent the elements 4. Evaluation of the relationships between the base visuals (A to B, A to C) through comparison of node characteristics. Creating directed links to represent the

  • Left Quote Icon

    Student Picture

  • Left Quote Icon

    Student Picture

  • Left Quote Icon

    Student Picture