Hypernetworks: A Molecular Evolutionary Architecture for Cognitive Learning and Memory

Abstract

Recent interest in human-level intelligence suggests a rethink of the role of machine learning in computational intelligence. We argue that without cognitive learning the goal of achieving human-level synthetic intelligence is far from completion. Here we review the principles underlying human learning and memory, and identify three of them, i.e., continuity, glocality, and compositionality, as the most fundamental to human-level machine learning. We then propose the recently-developed hypernetwork model as a candidate architecture for cognitive learning and memory. Hypernetworks are a random hypergraph structure higher-order probabilistic relations of data by an evolutionary self-organizing process based on molecular selfassembly. The chemically-based massive interaction for information organization and processing in the molecular hypernetworks, referred to as hyperinteractionism, is contrasted with the symbolist, connectionist, and dynamicist approaches to mind and intelligence. We demonstrate the generative learning capability of the hypernetworks to simulate linguistic recall memory, visual imagery, and language-vision crossmodal translation based on a video corpus of movies and dramas in a multimodal memory game environment. We also offer prospects for the hyperinteractionistic molecular mind approach to a unified theory of cognitive learning.

Layered Hypernetwork Models for Cross-Modal Associative Text and Image Keyword Generation in Multimodal Information Retrieval

“…The properties of the hypernetwork model are summarized as three aspects: glocality, compositionality and self association based on randomness and recall [5].…”

“…(8) More explanations on the derivative of the log-likelihood are showed in [5]. Therefore, log-likelihood of hypernetwork can be maximized by decreasing the difference of hyperedges from a given data set.…”

“…Learning of the first-layer hypernetworks is similar to the learning of conventional hypernetworks [4][5] except building a hypernetwork per one modality. At first, multimodal data are separated by modalities.…”

“…As a model to learn higher-order cross-modal associations, we used hypernetwork models in the previous study [4]. A hypernetwork is a higher-order probabilistic graphical model which has properties including glocality, compositionality, self-assembly, and recall-memory [5]. In the previous study, we showed that images could be retrieved with multimodal queries by text-to-image inference with trained hypernetworks [4].…”

MMG: A Learning Game Platform for Understanding and Predicting Human Recall Memory

“…These techniques include probabilistic graphical models, such as Bayesian networks [5], Markov random fields [6], and Markov logic networks [7]. More recent models include deep belief networks [8] and random hypergraph structures [9] for parallel associative memory.…”