Do LoRA Read Directions Encode Visual Concepts?

TLDRI compare the semantic coherence of read directions learned by standard, ReLU, and TopK LoRA adapters with random directions in CLIP’s residual stream. Clarity, a measure of semantic coherence, is concentrated at the positive and negative extremes of the activation distribution.Random directions can occasionally produce highly coherent examples, so a convincing activation grid alone does not show that a concept was learned. However, learned directions are more consistently coherent: 86% of s...