We provide an overview of different graph analytics approaches, with specific focus on a next generation graph with an associative memory-based reasoning system. As a form of latent or lazy learning, associative memories allow instant storage of new experiences, then reasoning by similarity to past experiences.
To implement this associative epistemology for scale and performance, these graphs are compressed, distributed, hyper-sparse matrices, where each matrix represents a 3600 view of each person, place, thing, situation, action, event, or outcome and how they are connected across multiple data sources and types. We will provide use case examples from insurance, tech, and intelligence industries, including KYC “segment of one” product recommendation, analytic/investigative workflow recommendation, fraud-waste-abuse, and issue/defect resolution related to IT and product quality management.
Although this memory-based approach is distinct from traditional machine learning and more recently deep learning, Complementary Learning Theory will be described as a way to unify models and memories, both required for leveraging a more comprehensive approach to ML/AI and more effectively augmenting data-driven decision-making. While Intel Saffron’s memory-matrix approach unifies graph semantics and numerical statistics, the future must also unify predicate semantics with associative semantics within the larger realm of AI Knowledge Representation and Reasoning.
