The ability to learn temporal relationships and use that knowledge to simulate future events is among the most remarkable aspects of cognition. Recently introduced behavioral task called Judgment of Imminence (JOI) combined with a well-known Judgment of Recency (JOR) task pointed to a remarkable symmetry between the temporal organization of memory and prediction. The data were consistent with the hypothesis that both memory and prediction can be organized as a compressed mental timeline. This means that the past and future can be remembered or simulated sequentially relative to the present. The compression implies that events closer to the present, regardless of whether they are in the past or in the future, were represented more accurately than those further from the present. Here we used the existing JOR model based on a compressed memory timeline to build an associative representation that can learn the temporal relationships and create a timeline of the future, which mirrors the timeline of the past. We show that this approach can simultaneously account for response times and accuracy in both JOR and JOI. This work provides a time-local neural-level mechanistic account for how the temporal organization of the memory can be used to learn the temporal structure of the world and simulate the future in an efficient manner as a compressed mental timeline.

Constrained by results from classic behavioral experiments weprovide a neural-level cognitive architecture for modeling be-havior in working memory tasks. We propose a canonicalmicrocircuit that can be used as a building block for work-ing memory, decision making and cognitive control. The con-troller controls gates to route the flow of information betweenthe working memory and the evidence accumulator and setsparameters of the circuits. We show that this type of cognitivearchitecture can account for results in behavioral experimentssuch as judgment of recency, probe recognition and delayed-match-to-sample. In addition, the neural dynamics generatedby the cognitive architecture provides a good match with neu-rophysiological data from rodents and monkeys. For instance,it generates cells tuned to a particular amount of elapsed time(time cells), to a particular position in space (place cells) andto a particular amount of accumulated evidence.

Hippocampal time cells carry a record of recent experience by firing during a circumscribed period of time after a triggeringstimulus. Different cells have “time fields’ at different delays up to at least tens of seconds. The observation that timefields representing events further in the past are wider supports the hypothesis that the more distant past is recorded withless resolution. However, previous studies have analyzed time fields averaged over trials, leading to the possibility thatthis could be a trial-averaging artifact. We analyzed single-unit recorded time cells with a hierarchical-Bayesian modelthat separately estimated within-trial receptive field width and across-trial variability. Even after isolating across-trialvariability, time-field width covaries with delay, consistent with less resolution for past events. Moreover, the shift oftime-fields for simultaneously recorded time cells correlates with each other, suggesting that time runs at different ratesfor the population from one trial to the next.

Cognitive psychologists have hypothesized that episodic recall is caused by the recovery of a gradually-changing state ofspatiotemporal context. Little is known about the processes that cause successful recovery of this temporal context. Recentbehavioral evidence suggests that in continuous recognition tasks, the retrieval time necessary to recover a previous contextdepends on the recency of the memory. Previous work has found that the non-decision time to retrieve a memory goes upwith the logarithm of its recency. This suggests retrieval of temporal context proceeds via scanning along a compressedtimeline but also contradicts earlier work suggesting that recency affects the drift rate of retrieval more than the non-decision time. Here we explore the effect of multiple repetitions on this counterintuitive result in continuous recognition.Our results find that while repeating items speeds up the time to access a memory, the recency effect persists out to at leastfive repetitions.

It is well known that, all things being equal, the accuracy ofmammalian timing and memory decays gradually with the pas-sage of time. The gradual decay of temporal accuracy is alsoobserved in single-unit neural recordings. Here we review re-cent modeling work describing a specific mechanism for tim-ing and memory and relevant neural data. The model describesa neural mechanism that can give rise to a logarithmically com-pressed representation of the recent past. We examine the spe-cific predictions of the model, in particular that the elapse oftime is represented by sequentially activated cells which firefor a circumscribed period of time. Such cells, called timecells, have been observed in neural recordings from severalbrain regions in multiple species. As predicted by the model,the cells show accuracy that decreases with time.

It is well-known that in free recall participants tend to recallwords presented close together in time in sequence, reflectinga form of temporal binding in memory. This contiguity effectis robust, having been observed across many different experi-mental manipulations. In order to explore a potential boundaryon the contiguity effect, participants performed a free recalltask in which items were presented at rates ranging from 2 Hzto 8 Hz. Participants were still able to recall items even atthe fastest presentation rate, though accuracy decreased. Im-portantly, the contiguity effect flattened as presentation ratesincreased. These findings illuminate possible constraints onthe temporal encoding of episodic memories.