Teaching AI to See the World More Like Humans Do

Introduces a three-step method for aligning AI visual representations with human knowledge: 1. Train a teacher model on the THINGS dataset using a small adapter to avoid overfitting. 2. Generate a large synthetic dataset (AligNet) of human-like odd-one-out decisions using the teacher model. 3. Fine-tune student AI models on AligNet to restructure their internal maps according to human conceptual hierarchies. Demonstrates improved human-alignment, robustness, generalization, and few-shot learning.