Neural network based successor representations to form cognitive maps of space and language

Abstract

How does the mind organize thoughts? The hippocampal-entorhinal complex is thought to support domain-general representation and processing of structural knowledge of arbitrary state, feature and concept spaces. In particular, it enables the formation of cognitive maps, and navigation on these maps, thereby broadly contributing to cognition. It has been proposed that the concept of multi-scale successor representations provides an explanation of the underlying computations performed by place and grid cells. Here, we present a neural network based approach to learn such representations, and its application to different scenarios: a spatial exploration task based on supervised learning, a spatial navigation task based on reinforcement learning, and a non-spatial task where linguistic constructions have to be inferred by observing sample sentences. In all scenarios, the neural network correctly learns and approximates the underlying structure by building successor representations. Furthermore, the resulting neural firing patterns are strikingly similar to experimentally observed place and grid cell firing patterns. We conclude that cognitive maps and neural network-based successor representations of structured knowledge provide a promising way to overcome some of the short comings of deep learning towards artificial general intelligence.

Classification at the accuracy limit: facing the problem of data ambiguity

“…This interesting finding, previously reported in Schilling et al 15 , suggests that unsupervised dimensionality reduction could be used to automatically detect and enhance natural clustering in unlabeled data. In combination with automatic labeling methods, such as Gaussian Mixture Models 41 , or the concept of the successor representations to map complex data structures 24 , 42 , this may provide an objective way to define ’natural kinds’ in arbitrary data sets.…”

“…By color-coding each projected data point of a data set according to its label, the representation of the data can be visualized as a set of point clusters. For instance, MDS has already been applied to visualize for instance word class distributions of different linguistic corpora 23 , hidden layer representations (embeddings) of artificial neural networks 6,15,24 , structure and dynamics of recurrent neural networks [25][26][27][28] , or brain activity patterns assessed during e.g. pure tone or speech perception 14,23,29 , or even during sleep 5,30 .…”

Bridging the Chasm Between Cognitive Representations and Formal Structures of Linguistic Meanings

“…This automatically rules out the contention that semantic inquiry in the two paradigms taps into distinct properties of linguistic meaning. On the empirical front, there is a good amount of neuroscientific evidence that suggests that semantic representations are often map-like or frame-like and such representations can be implemented by neural networks in the entorhinal-hippocampal complex (Covington & Duff 2016;Piai et al 2016;Stoewer et al, 2022). Such map-like representations are also organized around concepts of objects and things that form the basis of semantic representations in cortical columns (see Hawkins, Ahmad, & Cui, 2017;Lewis, Purdy, Ahmad, & Hawkins, 2019).…”

A Theory of Mental Frameworks

“…The more complex nature of problem identification is also born out in neuroscientific studies. Just as multiple-choice questions are easier to answer than open-ended questions, involving fewer complex networks in the brain (Zhang et al, 2021), the retrieval of information to answer a question is less neurologically complex than formulating one (Stoewer et al, 2022). As a teaching strategy, Rothstein and Santana encourage us to Make Just One Change: Teach Students to Ask Their Own Questions Rothstein and Santana (2011).…”