Computational constraints on the associative recall of spatial scenes

Abstract

We consider a model of associative storage and retrieval of compositional memories in an extended cortical network. Our model network is comprised of Potts units, which represent patches of cortex, interacting through long-range connections. The critical assumption is that a memory, for example of a spatial view, is composed of a limited number of items, each of which has a pre-established representation: storing a new memory only involves acquiring the connections, if novel, among the participating items. The model is shown to have a much lower storage capacity than when it stores simple unitary representations. It is also shown that an input from the hippocampus facilitates associative retrieval. When it is absent, it is advantageous to cue rare rather than frequent items. The implications of these results for emerging trends in empirical research are discussed.

Using multi-task experiments to test principles of hippocampal function

“…Human studies using high-resolution functional neuroimaging support this model by showing that neural activity in subregions of the hippocampus is subject to adaptation effects only upon repeated display of identical objects, rather than visually changed objects, supporting pattern separation in the human hippocampus and dentate gyrus (Bakker et al, 2008). Related work in this special issue explores computational constraints on associative memory built on associated, sparse concepts (Ryom et al, 2023). Collectively, this work is consistent with the hippocampus having pattern separation.…”

Hippocampal spatial view cells, place cells, and concept cells: View representations

“…Gastaldi et al (2021) then considered a number of statistical models (including that of Boboeva et al (2018) which had already dealt with correlations between the patterns stored in an associative memory), but they did not redo the capacity analysis nor the simulations with the correlations among all memories. If Gastaldi et al (2021) had done that (with simulations, because the analytical approach is difficult without the random correlation assumption of their equation 25), they would have found, subject to differences in the details, the same reduction in capacity that was reported by Ryom et al (2023) in their paper in this Special Issue of Hippocampus.…”

“…This is an important step in understanding how the hippocampus can help the neocortex to build semantic information, by recalling episodic memories to the neocortex for further processing (Rolls, 2022). Exactly how the neocortex builds its semantic memories with the different components in different cortical regions that are heavily interconnected as shown above and as illustrated in Figures 3 and 4 (Rolls et al, 2022a) is not clear, but models are being developed (Boboeva et al, 2018; Rolls, 2023a; Ryom et al, 2023; see also Sections 9 and 10).…”

“…In some parts of these regions, words are represented, and these word‐level representations may be important in helping semantic representations to be kept somewhat discrete, and not so overlapping in their components that there is a big impact on the neocortical memory capacity for semantic representations. That may be part of a solution to the issues about the representation of information in neocortical semantic memory systems when the memory patterns to be stored are somewhat correlated with each other that are raised by Treves and colleagues (Ryom et al, 2023; see Section 11). The point here is that words are arbitrary, relatively orthogonal, symbols, and may thus help to decorrelate neocortical semantic representations.…”

“…This is a commentary on issues important in understanding hippocampal function that are raised by papers in the Special Issue of Hippocampus (2023) entitled “Hippocampal system neurons encoding views in different species” (Alexander et al, 2023; Corrigan et al, 2023; Donoghue et al, 2023; LaChance & Taube, 2023; Lee et al, 2023; Quian Quiroga, 2023; Rolls, 2023b; Ryom et al, 2023; Wang et al, 2023; Wirth, 2023; Yang et al, 2023; Zhu et al, 2023).…”