The GraphLearner is a scalable neuromorphic algorithm capable of simulating high-order Markov Chains with efficient memory usage. Drawing inspiration from neocortical models, the GraphLearner addresses the challenges of traditional Markov Chain representations, which require exponential space for high orders. Utilizing Counting Bloom Filters, it achieves linear space complexity while preserving accuracy in sequence prediction tasks. This paper demonstrates how the GraphLearner replicates high-order Markov Chains, implements an oblivion mechanism to enhance robustness, and demonstrates that the benefits of this mechanism are preserved with increasing scales of data.