A model of egocentric to allocentric understanding in mammalian brains

Abstract

In the mammalian brain, allocentric representations support efficient self-location and flexible navigation. A number of distinct populations of these spatial responses have been identified but no unified function has been shown to account for their emergence. Here we developed a network, trained with a simple predictive objective, that was capable of mapping egocentric information into an allocentric spatial reference frame. The prediction of visual inputs was sufficient to drive the appearance of spatial representations resembling those observed in rodents: head direction, boundary vector, and place cells, along with the recently discovered egocentric boundary cells, suggesting predictive coding as a principle for their emergence in animals. The network learned a solution for head direction tracking convergent with known biological connectivity, while suggesting a possible mechanism of boundary cell remapping. Moreover, like mammalian representations, responses were robust to environmental manipulations, including exposure to novel settings, and could be replayed in the absence of perceptual input, providing the means for offline learning. In contrast to existing reinforcement learning approaches, agents equipped with this network were able to flexibly reuse learnt behaviours - adapting rapidly to unfamiliar environments. Thus, our results indicate that these representations, derived from a simple egocentric predictive framework, form an efficient basis-set for cognitive mapping.

State transitions in the statistically stable place cell population are determined by rate of perceptual change

“…Notably, these models focused on the description of individual fields but were agnostic of population-level interactions which featured in models that emphasised CA3 recurrence and mutual inhibition (Káli and Dayan, 2000;Samsonovich and McNaughton, 1997). Thus, our current results indicate that a synthesis of both approaches is important for understanding how the hippocampus represents large-scale spaces -yoking the evolution of statistically stable population-level activity to movement through visual states (Hedrick and Zhang, 2016;Uria et al, 2020).…”

“…In contrast, models based on attractor dynamics tend to ignore any systematic variance in place field size and density across environments, emphasising even coverage and carefully balanced activity (Káli and Dayan, 2000;Samsonovich and McNaughton, 1997). These two classes of model, as well as others (de Cothi and Stachenfeld et al, 2017;Uria et al, 2020), provide competing but not incompatible explanations of hippocampal dynamics but the evidence needed to generate a synthesis is lacking.…”

A connectome of theDrosophilacentral complex reveals network motifs suitable for flexible navigation and context-dependent action selection

“…Some units in these deep networks develop response properties broadly similar to those observed in real brains. However, insights into how such artificial neural networks generate the representations observed in their units-something that could, in principle, guide mechanistic hypotheses for the function of natural neural networks-have been slower to come (but see (Cueva et al, 2019;Uria et al, 2020) for progress in uncovering the architectural basis of navigational responses in these networks). In this era of deep learning, a broader question concerns the level of understanding that is appropriate or even possible for the function of large and complex neural networks (Gao and Ganguli, 2015;Hasson et al, 2020;Lillicrap and Kording, 2019;Richards et al, 2019;Saxe et al, 2020;Yamins and DiCarlo, 2016).…”

Robot navigation as hierarchical active inference

“…In recent work [65], a recurrent neural network was trained to predict sequences of visual inputs from (the latent space) of variational autoencoders.…”