Cortical reactivations predict future sensory responses

Abstract

SummaryPrevailing theories of offline memory consolidation posit that the pattern of neurons activated during a salient sensory experience will be faithfully reactivated, thereby stabilizing the entire pattern1-3. However, sensory-evoked patterns are not stable, but instead drift across repeated experiences4-7. To investigate potential roles of reactivations in the stabilization and/or drift of sensory representations, we imaged calcium activity of thousands of excitatory neurons in mouse lateral visual cortex. Presentation of a stimulus resulted in transient, stimulus-specific reactivations during the following minute. These reactivations depended on local circuit activity, as they were abolished by local silencing during the preceding stimulus. Contrary to prevailing theories, reactivations systemically differed from previous patterns evoked by the stimulus. Instead, they were more similar to future patterns evoked by the stimulus, therebypredictingrepresentational drift. In particular, neurons that participated more or less in early reactivations than in stimulus response patterns subsequently increased or decreased their future stimulus responses, respectively. The rate and content of these reactivations was sufficient to accurately predict future changes in stimulus responses and, surprisingly, the decreasing similarity of responses to distinct stimuli. Thus, activity patterns during sensory cortical reactivations may guide the drift in sensory responses to improve sensory discrimination8.

Reward timescale controls the rate of behavioral and dopaminergic learning

“…One possibility by which "trial-based" frameworks of learning could be used to account for the data presented here is to assume that replays/reactivations of cue-reward experiences during the extended ITI provides "virtual trials" in lieu of the real trials experienced by mice 56,59 . While we cannot rule out this possibility, the similarity between the learning curves for the long and short ITI groups suggests that such a mechanism would somehow have to precisely replicate the effect of the missing 9/10 experiences in the long ITI group, and the available evidence suggests that neither the structure of replay/reactivation events nor the information they encode perfectly replicate past experiences [60][61][62][63][64][65][66] . We therefore suggest that our proposed non-trial-based learning rule is a more parsimonious explanation of the quantitative scaling observed here.…”

“…However, most demonstrations of the effectiveness of spaced versus massed cue-reward learning examine massed learning with a short ITI relative to the cue-reward delay 27,32 (ITI/trial ratio less than 10). Demonstrations like these do not rule out TDRL because under these conditions, extensions of TDRL that account for the ITI 37 can produce faster learning with longer ITIs (Extended Data Fig 1,60 s vs. 6 s). Thus, to rigorously examine the predictions of the dominant neurobiological "trial-based" models, experiments testing categorical, falsifiable predictions shared by TDRL implementations should be identified.…”

Representational drift as a result of implicit regularization

“…Experimentally, ( Deitch et al, 2021 ) reported a decrease in activity at the beginning of the experiment, which they suggested was correlated with some behavioral change, but we believe it could also be a result of the directed drift phase. ( Nguyen et al, 2022 ) also reported a slow directed change in representation long after familiarity with the stimuli. There is another consequence of the timescale separation.…”

Representational drift as a result of implicit regularization

“…Experimentally, [7] reported a decrease in activity at the beginning of the experiment, which they suggested was correlated with some behavioral change, but we believe it could also be a result of the directed drift phase. [40] also reported a slow directed change in representation long after familiarity with the stimuli. There is another consequence of the timescale separation.…”