How the brain stores and retrieves memories is an important unsolved problem in neuroscience. It is commonly believed that memories are represented in the brain by distinct patterns of neural activity. Attractor dynamics has been proposed as one of the theoretical frameworks for learning and memory in neural networks. However, most of the prior theoretical work typically assumes that the neural network consists of fully-connected, binary neurons and that neuronal representations of memories are uncorrelated. In this paper, we propose a model consisting of continuously varying, rate-based, sparse neural network with a local learning rule which stores correlated patterns organized into multiple uncorrelated classes. We perform analytical calculations to compute maximum storage capacity and basin of attraction. It is found that increasing pattern correlation decreases storage capacity, and increasing the memory load decreases basin of attraction. We also study rate-based and spiking based neural network with separate excitatory and inhibitory populations and tight EI balance. Recent experimental work indicates that piriform cortex exhibits attractor dynamics and possibly stores hierarchically correlated patterns. So, we consider this work as a model for olfactory memory storage.