We consider the dictionary learning problem, where the aim is to model the given data as a linear combination of a few columns of a matrix known as a dictionary, where the sparse weights forming the linear combination are known as coefficients. Since both the dictionary and coefficients parameterizing the linear model are unknown, the corresponding optimization is inherently non-convex... This was a major challenge until recently, when provable algorithms for dictionary learning were proposed. Yet, these provide guarantees only on the recovery of the dictionary, without explicit recovery guarantees on the coefficients. Moreover, any estimation error in the dictionary adversely impacts the ability to successfully localize and estimate the coefficients. This potentially limits the utility of existing provable dictionary learning methods in applications where coefficient recovery is of interest. To this end, we develop a simple online alternating optimization-based algorithm for dictionary learning, which recovers both the dictionary and coefficients exactly at a geometric rate. Specifically, we show that -- when initialized appropriately -- the algorithm linearly converges to the true factors. Our algorithm is also scalable and amenable for large scale distributed implementations in neural architectures, by which we mean that it only involves simple linear and non-linear operations. Finally, we corroborate these theoretical results via experimental evaluation of the proposed algorithm with the current state-of-the-art techniques. (read more)