Learning, Fast and Slow: Single- and Many-Shot Learning in the Hippocampus

Abstract

The hippocampus is critical for memory and spatial navigation. The ability to map novel environments, as well as more abstract conceptual relationships, is fundamental to the cognitive flexibility that humans and other animals require to survive in a dynamic world. In this review, we survey recent advances in our understanding of how this flexibility is implemented anatomically and functionally by hippocampal circuitry, during both active exploration (online) and rest (offline). We discuss the advantages and limitations of spike timing-dependent plasticity and the more recently discovered behavioral timescale synaptic plasticity in supporting distinct learning modes in the hippocampus. Finally, we suggest complementary roles for these plasticity types in explaining many-shot and single-shot learning in the hippocampus and discuss how these rules could work together to support the learning of cognitive maps.

Keywords: hippocampus; learning; memory; plasticity.

Figure 1

Two views of the hippocampus, (a) Classic anatomical view of the hippocampus. Panel a adapted with permission from Neves et al. (2008). (b) Functional view of the hippocampus. Figure adapted from images created with BioRender.com.

Figure 2

Dendritic tuning supports single-shot learning via behavioral timescale synaptic plasticity (BTSP). (a) Schematic of dendritic and somatic tuning. (b) Inputs to cell shown in panel a. Heights represent the weight of each input. (c) Output of the cell. (d) Color-coded heatmap of spatial firing rates of a CA1 pyramidal cell showing place field emergence after a dendritic plateau (top); somatic membrane potential (gray); and low-pass filtered membrane potential (blue) for laps preceding (lap 10), during (lap 11), and after plateau (lap 27) (bottom). (e) Heterogeneous spatial tuning properties of small-diameter dendrites of CA1 pyramidal cells. Tuning curve correlations are calculated exclusively between connected, coimaged dendrites (to left of each bar chart). Vertical histograms show distributions of minimum circular distances between somatic and dendritic place field centers. Bar widths represent the relative abundance of bin values. Panel d adapted with permission from Bittner et al. (2017), and panel e adapted with permission from O’Hare et al. (2022). Panels a–c adapted from images created with BioRender.com.

Figure 3

Online and offline hippocampal sequences. (a) Place cell activity during traversal (left) and rest (right). (b) Zoom in showing sequence activity during online theta (left) and replayed during offline sharp-wave ripples (SWRs) (right). (c) Pairwise sequential relationships are acquired online via Hebbian plasticity and read out during offline replay. (d) Online place cell sequences recorded with two-photon calcium imaging in the hippocampal area CA1. An example of three run laps on a circular treadmill belt shows (i) that the position of the animal (blue line) could be decoded (pink line) using (ii) the deconvolved calcium activity of the 180 pyramidal cells (PCs) imaged during this session. (iii) The wavelet spectrogram of the CA1 local field potential (LFP) shows increases in theta-band power during locomotion. Example of an online place cell sequence, arranged by the cells’ peak firing location on a treadmill belt (iv). Two example population synchrony events occurring during immobility with significant forward (top) or reverse (bottom) replay are shown, including the raw and SWR-filtered LFP traces (gray and red lines), the raster of the participating PCs sorted by place field position and the Bayesian decoded posterior probability of position (tiled twice to account for reward-spanning trajectories) (v). Panel d adapted from Grosmark et al. (2021).

Figure 4

Hebbian sequence learning, and a possible role for inhibitory plasticity. (a) Schematic of a random cue (RC) task, where mice are presented with sensory cues at random locations and random intertrial intervals (ITIs). (b) Average peri-sharp-wave ripple (peri-SWR) fluorescence (ΔF/F₀) response of axons from cue-responsive CA3 Schaffer collateral (CA3SC) axons during the RC task. Note the gradual suppression of the response magnitude. (c) Average peri-SWR responses for cue-responsive (red) and other (gray) CA3SC axons during stimulus-free periods before and after RC sensory stimulation. Note the selective suppression of cue-responsive axons during SWRs and during stimulation and after. (d) Latent sequences in the world, with noisy observations. (e) Hebbian plasticity embeds sequentially observed pairs into a weight matrix (black +). This also results in the learning of spurious associations (red +). Inhibitory plasticity (bottom) prevents the replay of distractor stimuli. (f) Population activity during online exploration and offline replay. Nongeneralizable stimuli that are represented online are selectively suppressed from consolidation. Panels a–c adapted from Terada et al. (2022), and panels d–f adapted from images created with BioRender.com and previously published in Liao et al. (2022).