project3_kbai.pdf - 1 Project 3 CS7637 Sukeerthi Varadarajan [email protected] Introduction Raven\u2019s Progressive Matrices(RPM is an intelligence

# project3_kbai.pdf - 1 Project 3 CS7637 Sukeerthi...

• 8
• 100% (11) 11 out of 11 people found this document helpful

This preview shows page 1 - 3 out of 8 pages.

1 Project 3 : CS 7637 Sukeerthi Varadarajan Introduction Raven’s Progressive Matrices (RPM) is an intelligence test that is used to measure abstract reasoning across all age groups, ordered by increasing difficulty. In this project, we are to create an AI agent that solves Ravens Progressive Matrices (RPM) – problem sets B, C, D and E. We will take a hybrid approach that uses image (for D and E) and text based (for B and C) Approach The agent reasons over the problem it receives by representing all the objects in the matrix and the answer figures verbally in a Semantic Network. Then the agent uses Generate and Test method to loop through each of the options, placing an option in place of figure D (for 2x2 matrices) or I (for 3x3 matrices). The agent then compares the relationships and transformations with the other figures, i.e. from A to H with I (for problem set C). The agent also compare between the question figures, i.e. A to B and A to C and computes different types of transformations such as vertical, horizontal and diagonal. Each of the transformations has weights assigned to them and the agent uses similarity metrics to score each of the options and the transformations performed are stored in a dictionary. The option with the highest similarity score is chosen as the answer. At the end of Project 2, we had the following transformations- similarity, rotation, reflection, alignment, fill, deletion, shape, angle, size, overlap and positional. We will add more visual transformations based on the performance of the agent with the D and E problems.
2 Journal Entry 1 Submission Date – April 19, 2019- 03:39:02 UTC For this submission, I submitted the final agent as was obtained at the end of project 2. No changes were made to the code or approach. The reason why I submitted the same code was because the agent could have solved some of the examples in sets D and E based on the transformations described. Results The verbal agent is able to identify the correct answers for some cases like- Basic E- 06 and Basic D-05 since the agent is able to derive the answer by just comparing A and C and G (in the case of Basic D-05) and one of the answer options and the basic transformations are able to capture the transformations between the question images and the answer- such as the presence of similarity transforms between the question images in Basic E-06.