Visual Sequences IQ Tests Tags Search Pattern Recognition

  • Slides: 14
Download presentation
Visual Sequences IQ Tests Tags : Search, Pattern Recognition, Logic etc Dipendra Kumar Misra

Visual Sequences IQ Tests Tags : Search, Pattern Recognition, Logic etc Dipendra Kumar Misra (Y 9201) Mukul Singh (Y 9350) Advisor : Dr. Amitabh Mukherjee Dr. Sumit Gulwani 1

Sample Problem Which Image logically follows the Input Sequence 2

Sample Problem Which Image logically follows the Input Sequence 2

Motivation • Automate Manual Verification of Intelligence Problems • Synthesis of IQ Problems •

Motivation • Automate Manual Verification of Intelligence Problems • Synthesis of IQ Problems • Insights into cognitive studies • Visual Problems combine both Logic and Geometry 3

Background Work • IQ problems have been studied for a long time • Fanya

Background Work • IQ problems have been studied for a long time • Fanya Montalvo [1986, 1993] discusses topics such as Diagram Understanding • “It will be hard for a computer program to solve picture problems” – Sanghi & Dowe 4

Case Study • No Formal Definition of IQ problems • Separate Semantic Based problems

Case Study • No Formal Definition of IQ problems • Separate Semantic Based problems from Syntax Based problems • Case Study on over 100 problems from standard books and online tests 5

Our Intuition “A Transformation Language with few transformations like Rotation and Shading should be

Our Intuition “A Transformation Language with few transformations like Rotation and Shading should be powerful enough to cover a large subset of problems” 6

Methodology Set of Images ( Test Set and Solution Set ) Front End Back

Methodology Set of Images ( Test Set and Solution Set ) Front End Back End Solution 7

Front End Set of Images (Test Set and solution set) L 1 Set of

Front End Set of Images (Test Set and solution set) L 1 Set of Lines, Circles and Special Objects detected in the image set L 2 Polygons formed by combining suitable lines 8

Back End P 1 T 2 T 1 P 2 T 3 P 4

Back End P 1 T 2 T 1 P 2 T 3 P 4 P 3 T 4 P 5 T 5 P 6 Transformation of P 1 = T 1 ∩ T 3 U T 2 ∩ (T 4 U T 5) 9

Ranking Method Rank(P, Q) = Match. Primitive(P, Q) + Match. Primitive(Q, P) Match. Primitive(P,

Ranking Method Rank(P, Q) = Match. Primitive(P, Q) + Match. Primitive(Q, P) Match. Primitive(P, Q) = Number of Primitive of P in Q 10

Lets Do Some Practice 11

Lets Do Some Practice 11

Handling Ambiguity • Multiple Transformations – • Use Least Complex Formula (Kolmogorov) P R

Handling Ambiguity • Multiple Transformations – • Use Least Complex Formula (Kolmogorov) P R 90 A Q R 270 B R 90 R 270 A B Choosing Cross Terms (90, 270) will give us wrong answer 12

Results • Solve Visual Problems for a Large Subset • Zero Human Intervention •

Results • Solve Visual Problems for a Large Subset • Zero Human Intervention • No restriction on type of primitives – apple, cap, chair etc. 13

References & Resources • “IQ Tests to Keep You Sharp” by Philip Carter and

References & Resources • “IQ Tests to Keep You Sharp” by Philip Carter and Kenneth Russel • “Check Your IQ” by Ken Russell and Philip Carter • www. similarminds. com • Spreadsheet Data Manipulation Using Examples S. Gulwani POPL 2011 • A computer program capable of passing IQ tests, Sanghi, P. and Dowe 2003 • Montalvo, Fanya S. Diagram understanding, 1986 14