Generative Search Model Compares With Human Vision
Khajuria, Tulver, and Aru (published Dec 9, 2025) implement a generative search algorithm using genetic search to solve a challenging constellation-based visual problem and compare its performance to human participants. They measure accuracy, reaction time, and drawing overlap, reporting that generative analysis-by-synthesis can emulate aspects of human iterative problem solving and inform design of more robust vision systems.
Key Points
- 1Implements genetic-search-based analysis-by-synthesis to infer object contours in noisy constellation-style images.
- 2Demonstrates comparable behaviors to humans in accuracy, reaction time, and drawing overlap on the task.
- 3Suggests generative search can guide robust, iterative visual inference and inform design of perceptually resilient vision systems.
Scoring Rationale
Combines peer-reviewed empirical comparison and usable code; demonstrates strong results but limited novelty beyond specialized constellation task.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
