# Project 1 Reflection.pdf - 1 Project 1 Reflection...

• Homework Help
• 6
• 91% (11) 10 out of 11 people found this document helpful

This preview shows page 1 - 3 out of 6 pages.

1 Project 1 ReflectionIntroductionRaven’s Progressive Matrices (RPM) are a human intelligence test that uses visualanalogies of geometric shapes to solve the missing figure within a problem matrix.For this project I will reflect on an AI agent created to solve problems similar tothose of Raven’s Progressive Matrices (RPM). When working with a RPM, thebottom right frame will be inferred from the contents of the rest of the matrix offrames. For example, Figure 1 below shows a 2x2 matrix, where analogies can bemade between figures A -> B and C -> answer (which I will call D), and between A-> C and B -> D.Figure 1: A 2x2 RPM Problem Project LimitationsI used a method tailored to the 2x2 problem mentioned above, using an entirelyverbal approach, where a human first describes the contents of each frame in aprecise and predictable manner, and that description is used to solve the problem.The catch is the verbal approach is more limited; it requires a human to manuallyspecify every object and input the necessary attributes and for the number ofattributes to be small enough that each one can be explicitly accounted for in the
2 programming (which means it can’t generalize where more attributes are required).My agent will simply skip any problem it wasn’t prepared to handle, like 3x3 orvisual only problems, and it would simply discard information for attributes itwasn’t explicitly built to handle. These tradeoffs will allow me to create a simplerand more robust agent for the cases the agent intends to solve; the agent could usedeterministic approaches whose logic could be grasped quickly, not needing to getinto image processing and the uncertainty it entails.Agent’s Problem SolvingAs mentioned, since the agent uses the verbal approach it does not read in the actualimages of the problem but rather the verbal description of each image which itstores in frames. The agent will use frames to provide easy access, reference, andupdates for comparisons and to gather information quickly for each image andshape. After the creation of the Semantic Network the agent is faced with threeproblems before it can choose an answer:Determine which shape in one image corresponds to a node in the secondimage.