Passive odor exposure can change olfactory perception beyond simple pattern separation. Recent studies reveal that principal neurons in the mouse olfactory bulb exhibit bi-directional changes in population- level representations that depend on input similarity. Our analysis reveals the presence of a systematic drift, manifested as rotation of population vectors away from their initial representations. We explain these phenomena using a recurrent inhibition feedback model with unsupervised Hebbian learning. Our model predicts a power- law decay in mean population activity over days of odor exposure, matching experimental observations. The basic network model accounts for decorrelation dynamics in similar odor inputs, while incorporating structural constraints reproduces both between-day rotational drift and bi-directional plasticity in both similar and dissimilar input scenarios. This work provides mechanistic insights into experience-dependent encoding changes in early olfactory processing and suggests that implicit perceptual learning in bulbar circuits can drive representational drift that influences cortical processing.