GraphLearner: An Approach to Sequence Recognition and Generation
Tim G. Harrison, Thomas J. Böhme, Mario Kubek, Herwig Unger · 2024
This paper presents GraphLearner a neuromorphic sequence generator with similarities to Markov Chain Models. GraphLearner is proposed as an alternative to ‘black box’ deep neural network models which lack explainability and adaptability. Bloom Filters are used to simplify otherwise computationally expensive Markov chain probability calculations. It is demonstrated with Natural Language Processing tasks, generating sentences of remarkable quality.