Utilization of generative AI for the characterization and identification of visual unknowns
Kara Combs, Trevor Bihl, Subhashini Ganapathy
Abstract
Current state-of-the-art artificial intelligence (AI) struggles with accurate interpretation of out-of-library objects. One method proposed remedy is analogical reasoning (AR), which utilizes abductive reasoning to draw inferences on an unfamiliar scenario given knowledge about a similar familiar scenario. Currently, applications of visual AR gravitate toward analogy-formatted image problems rather than real-world computer vision data sets. This paper proposes the Image Recognition Through Analogical Reasoning Algorithm (IRTARA) and its “generative AI” version called “GIRTARA” which describes and predicts out-of-library visual objects. IRTARA characterizes the out-of-library object through a list of words called the “term frequency list”. GIRTARA uses the term frequency list to predict what the out-of-library object is. To evaluate the quality of the results of IRTARA, both quantitative and qualitative assessments are used, including a baseline to compare the automated methods with human-generated results. The accuracy of GIRTARA’s predictions is calculated through a cosine similarity analysis. This study observed that IRTARA had consistent results in the term frequency list based on the three evaluation methods for the high-quality results and GIRTARA was able to obtain up to 65% match in terms of cosine similarity when compared to the out-of-library object’s true labels.