Decade of Knowledge-Guided Machine Learning in Computer Vision

A new systematic review maps how “knowledge-guided machine learning” (KGML) is reshaping computer vision after years of relying on data-only deep networks. Led by Christine Dewi at Satya Wacana Christian University with co-authors Dhananjay Thiruvady and Nayyar Zaidi of Deakin University, the open-access study—published in Multimedia Tools and Applications—analyzes research from January 2014 to July 2025. Following PRISMA screening, it reviews 286 high-quality studies out of 2,788 candidates. The authors argue that models trained purely on data can fail through bias, adversarial attacks, sensitive information leakage, overconfidence on incomplete inputs, and poor generalization. KGML addresses this by embedding structured scientific and domain knowledge into training.