Generative Predictive Codes by Multiplexed Hippocampal Neuronal Tuplets

Abstract

Rapid internal representations are continuously formed based on single experiential episodes in space and time, but the neuronal ensemble mechanisms enabling rapid encoding without constraining the capacity for multiple distinct representations are unknown. We developed a probabilistic statistical model of hippocampal spontaneous sequential activity and revealed existence of an internal model of generative predictive codes for the regularities of multiple future novel spatial sequences. During navigation, the inferred difference between external stimuli and the internal model was encoded by emergence of intrinsic-unlikely, novel functional connections, which updated the model by preferentially potentiating post-experience. This internal model and these predictive codes depended on neuronal organization into inferred modules of short, high-repeat sequential neuronal "tuplets" operating as "neuro-codons." We propose that flexible multiplexing of neuronal tuplets into repertoires of extended sequences vastly expands the capacity of hippocampal predictive codes, which could initiate top-down hierarchical cortical loops for spatial and mental navigation and rapid learning.

Content-based dissociation of hippocampal involvement in prediction

“…That is, upon presentation of a predictive cue or context, the hippocampus may retrieve the associated outcome through pattern completion (14,21), regardless of the exact nature of the stimuli. This is in line with evidence that the hippocampus is involved in many different types of predictions, pertaining to, for example, faces and scenes (22), auditory sequences (23), odors (24), and spatial locations (19,25).…”

A Model for Navigation in Unknown Environments Based on a Reservoir of Hippocampal Sequences

“…First, the place fields tend to get larger further into the linear track (Gothard et al 1996;Haas et al 2019), indicating that fields late in a sequence accumulate more sensory information. Second, Liu et al (2018a) found evidence for prewired sequence motifs being shorter than the observed sequences. The length of the sequences in the model of Appendix A varies with mean synaptic weights and as a function of all other parameters defining neuronal excitability, hence, arguing that transitions from short to long sequences can result from general excitability increases.…”

“…A potential criticism to the model arises from physiological reports that offline sequences are plastic and change from preplay to replay mostly by adding new neurons (Grosmark and Buzsaki 2016;Liu et al 2018a;Chenani et al 2019), whereas in the present model the sequences remain fixed. There are two ways to explain sequence plasticity in the realm of this model.…”

“…Touretzky and Redish 1996;Samsonovich and McNaughton 1997;Cutsuridis and Hasselmo 2012;Saravanan et al 2015;Erdem et al 2015;Stachenfeld et al 2017), however, there is still ongoing debate particularly on the mechanisms (e.g. Foster 2017; Liu et al 2018b;Matheus Gauy et al 2018;Nicola and Clopath 2019) and the functional role of temporal activity patterns (Liu et al 2018a). In contrast to technical systems, where autonomous localization methods are based on a costly collection of extensive series of sensory snapshots (Davison et al 2007;Milford and Wyeth 2012;Siam and Zhang 2017), the prevalent idea in neuroscience is that, in mammals, the hippocampal space code arises from efficient internal dynamics that is locked to places of salient sensory features (Keinath et al 2018) thereby saving the synaptic resources for explicitly memorizing all details of place-specific sensory inputs.…”

Replay as wavefronts and theta sequences as bump oscillations in a grid cell attractor network

“…Place cell activity appears to contain an abundance of certain short sequences that represent contiguous locations in novel environments and participate frequently in replays [112]. This finding is related to the disputed [113] observation of hippocampal preplay [114][115][116].…”