A theory of hippocampal function: New developments.

We develop further here the only quantitative theory of the storage of information in the hippocampal episodic memory system and its recall back to the neocortex. The theory is upgraded to account for a revolution in understanding of spatial representations in the primate, including human, hippocampus, that go beyond the place where the individual is located, to the location being viewed in a scene. This is fundamental to much primate episodic memory and navigation: functions supported in humans by pathways that build 'where' spatial view representations by feature combinations in a ventromedial visual cortical stream, separate from those for 'what' object and face information to the inferior temporal visual cortex, and for reward information from the orbitofrontal cortex. Key new computational developments include the capacity of the CA3 attractor network for storing whole charts of space; how the correlations inherent in self-organizing continuous spatial representations impact the storage capacity; how the CA3 network can combine continuous spatial and discrete object and reward representations; the roles of the rewards that reach the hippocampus in the later consolidation into long-term memory in part via cholinergic pathways from the orbitofrontal cortex; and new ways of analysing neocortical information storage using Potts networks.

Taking time to compose thoughts with prefrontal schemata.

Under what conditions can prefrontal cortex direct the composition of brain states, to generate coherent streams of thoughts? Using a simplified Potts model of cortical dynamics, crudely differentiated into two halves, we show that once activity levels are regulated, so as to disambiguate a single temporal sequence, whether the contents of the sequence are mainly determined by the frontal or by the posterior half, or by neither, depends on statistical parameters that describe its microcircuits. The frontal cortex tends to lead if it has more local attractors, longer lasting and stronger ones, in order of increasing importance. Its guidance is particularly effective to the extent that posterior cortices do not tend to transition from state to state on their own. The result may be related to prefrontal cortex enforcing its temporally-oriented schemata driving coherent sequences of brain states, unlike the atemporal "context" contributed by the hippocampus. Modelling a mild prefrontal (vs. posterior) lesion offers an account of mind-wandering and event construction deficits observed in prefrontal patients.

Déjà vu: A botched memory operation, illegitimate to start with.

Rather than a natural product, a computational analysis leads us to characterize déjà vu as a failure of memory retrieval, linked to the activation in neocortex of familiar items from a compositional memory in the absence of hippocampal input, and to a misappropriation by the self of what is of others.

Assessments of dentate gyrus function: discoveries and debates.

There has been considerable speculation regarding the function of the dentate gyrus (DG) - a subregion of the mammalian hippocampus - in learning and memory. In this Perspective article, we compare leading theories of DG function. We note that these theories all critically rely on the generation of distinct patterns of activity in the region to signal differences between experiences and to reduce interference between memories. However, these theories are divided by the roles they attribute to the DG during learning and recall and by the contributions they ascribe to specific inputs or cell types within the DG. These differences influence the information that the DG is thought to impart to downstream structures. We work towards a holistic view of the role of DG in learning and memory by first developing three critical questions to foster a dialogue between the leading theories. We then evaluate the extent to which previous studies address our questions, highlight remaining areas of conflict, and suggest future experiments to bridge these theories.

Narratives need not end well; nor say it all.

To fully embrace situations of radical uncertainty, we argue that the theory should abandon the requirements that narratives, in general, must lead to affective evaluation, and that they have to explain (and potentially simulate) all or even the bulk of the current decisional context. Evidence from studies of incidental learning show that narrative schemata can bias decisions while remaining fragmentary, insufficient for prediction, and devoid of utility values.

Computational constraints on the associative recall of spatial scenes.

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.

Challenges for Place and Grid Cell Models.

This chapter gives a short overview of computational models dealing with two fundamental building blocks in spatial cognition: grid and place cells, and of the open issues such models may help address.

Angular and linear speed cells in the parahippocampal circuits.

An essential role of the hippocampal region is to integrate information to compute and update representations. How this transpires is highly debated. Many theories hinge on the integration of self-motion signals and the existence of continuous attractor networks (CAN). CAN models hypothesise that neurons coding for navigational correlates - such as position and direction - receive inputs from cells conjunctively coding for position, direction, and self-motion. As yet, very little data exist on such conjunctive coding in the hippocampal region. Here, we report neurons coding for angular and linear velocity, uniformly distributed across the medial entorhinal cortex (MEC), the presubiculum and the parasubiculum, except for MEC layer II. Self-motion neurons often conjunctively encoded position and/or direction, yet lacked a structured organisation. These results offer insights as to how linear/angular speed - derivative in time of position/direction - may allow the updating of spatial representations, possibly uncovering a generalised algorithm to update any representation.

Has the hippocampus really forgotten about space?

Several lines of evidence, including the discovery of place cells, have contributed to the notion that the hippocampus serves primarily to navigate the environment, as a repository of spatial memories, like a drawer full of charts; and that in some species it has exapted on this original one an episodic memory function. We argue that recent evidence questions the primacy of space, and points at memory load, whether spatial or not, as the challenge that mammalian hippocampal circuitry has evolved to meet.

Navigating through the ebbs and flows of language.

Is progress in understanding the neural basis for spatial navigation relevant to the human language faculty? Not so much at the shortest scale, where movement is continuous, a recent study in the space of vowels suggests. At a much larger scale, however, that of the verbalization of run-away thoughts, a rich phenomenology appears to involve critical contributions by some of the brain structures also involved in spatial cognition. Their interactions may have to be approached with models operating at an integrated cortical level and allowing for the compositionality of multiple local attractor states. A useful window on the latching dynamics enabled by cortico-cortical interactions may be offered by altered states of consciousness. As an example, psychedelic states have been reported to alter the graph properties of functional connectivity in the cortex so as to facilitate wide-ranging trips.

Hyper-alignment: Great mice think alike.

Can the neural activity expressing the same mental processes in two different individuals be somehow aligned? Recent evidence suggests that in some cases it can, in mice, at least when they think about space, but possibly even when conjuring up something more abstract.

Continuous attractors for dynamic memories.

Episodic memory has a dynamic nature: when we recall past episodes, we retrieve not only their content, but also their temporal structure. The phenomenon of replay, in the hippocampus of mammals, offers a remarkable example of this temporal dynamics. However, most quantitative models of memory treat memories as static configurations, neglecting the temporal unfolding of the retrieval process. Here, we introduce a continuous attractor network model with a memory-dependent asymmetric component in the synaptic connectivity, which spontaneously breaks the equilibrium of the memory configurations and produces dynamic retrieval. The detailed analysis of the model with analytical calculations and numerical simulations shows that it can robustly retrieve multiple dynamical memories, and that this feature is largely independent of the details of its implementation. By calculating the storage capacity, we show that the dynamic component does not impair memory capacity, and can even enhance it in certain regimes.

When we recall a past experience, accessing what is known as an 'episodic memory', it usually does not appear as a still image or a snapshot of what occurred. Instead, our memories tend to be dynamic: we remember how a sequence of events unfolded, and when we do this, we often re-experience at least part of that same sequence. If the memory includes physical movement, the sequence combines space and time to remember a trajectory. For example, a mouse might remember how it went down a hole and found cheese there. However, mathematical models of how past experiences are stored in our brains and retrieved when we remember them have so far focused on snapshot memories. 'Attractor network models' are one type of mathematical model that neuroscientists use to represent how neurons communicate with each other to store memories. These models can provide insights into how circuits of neurons, for example those in the hippocampus (a part of the brain crucial for memory), may have evolved to remember the past, but so far they have only focused on how single moments, rather than sequences of events, are represented by populations of neurons. Spalla et al. found a way to extend these models, so they could analyse how networks of neurons can store and retrieve dynamic memories. These memories are represented in the brain as 'continuous attractors', which can be thought of as arrows that attract mental trajectories first to the arrow itself, and once on the arrow, to the arrowhead. Each recalled event elicits the next one on the arrow, as the mental trajectory advances towards the arrowhead. Spalla et al. determined that memory networks in the hippocampus of mammals can store large numbers of these 'arrows', up to the same amount of 'snapshot' memories predicted to be stored with similar models. Spalla et al.'s results may allow researchers to better understand memory storage and recall, since they allow for the modelling of complex and realistic aspects of episodic memories. This could provide insights into processes such as why our minds wander, as well as having implications for the study of how neurons physically interact with each other to transmit information.

Latching dynamics as a basis for short-term recall.

We discuss simple models for the transient storage in short-term memory of cortical patterns of activity, all based on the notion that their recall exploits the natural tendency of the cortex to hop from state to state-latching dynamics. We show that in one such model, and in simple spatial memory tasks we have given to human subjects, short-term memory can be limited to similar low capacity by interference effects, in tasks terminated by errors, and can exhibit similar sublinear scaling, when errors are overlooked. The same mechanism can drive serial recall if combined with weak order-encoding plasticity. Finally, even when storing randomly correlated patterns of activity the network demonstrates correlation-driven latching waves, which are reflected at the outer extremes of pattern space.

Efficiency of Local Learning Rules in Threshold-Linear Associative Networks.

We derive the Gardner storage capacity for associative networks of threshold linear units, and show that with Hebbian learning they can operate closer to such Gardner bound than binary networks, and even surpass it. This is largely achieved through a sparsification of the retrieved patterns, which we analyze for theoretical and empirical distributions of activity. As reaching the optimal capacity via nonlocal learning rules like back propagation requires slow and neurally implausible training procedures, our results indicate that one-shot self-organized Hebbian learning can be just as efficient.

In poetry, if meter has to help memory, it takes its time.

To test the idea that poetic meter emerged as a cognitive schema to aid verbal memory, we focused on classical Italian poetry and on three components of meter: rhyme, accent, and verse length. Meaningless poems were generated by introducing prosody-invariant non-words into passages from Dante's Divina Commedia and Ariosto's Orlando Furioso. We then ablated rhymes, modified accent patterns, or altered the number of syllables. The resulting versions of each non-poem were presented to Italian native speakers, who were then asked to retrieve three target non-words. Surprisingly, we found that the integrity of Dante's meter has no significant effect on memory performance. With Ariosto, instead, removing each component downgrades memory proportionally to its contribution to perceived metric plausibility. Counterintuitively, the fully metric versions required longer reaction times, implying that activating metric schemata involves a cognitive cost. Within schema theories, this finding provides evidence for high-level interactions between procedural and episodic memory.

Professional or Amateur? The Phonological Output Buffer as a Working Memory Operator.

The Phonological Output Buffer (POB) is thought to be the stage in language production where phonemes are held in working memory and assembled into words. The neural implementation of the POB remains unclear despite a wealth of phenomenological data. Individuals with POB impairment make phonological errors when they produce words and non-words, including phoneme omissions, insertions, transpositions, substitutions and perseverations. Errors can apply to different kinds and sizes of units, such as phonemes, number words, morphological affixes, and function words, and evidence from POB impairments suggests that units tend to substituted with units of the same kind-e.g., numbers with numbers and whole morphological affixes with other affixes. This suggests that different units are processed and stored in the POB in the same stage, but perhaps separately in different mini-stores. Further, similar impairments can affect the buffer used to produce Sign Language, which raises the question of whether it is instantiated in a distinct device with the same design. However, what appear as separate buffers may be distinct regions in the activity space of a single extended POB network, connected with a lexicon network. The self-consistency of this idea can be assessed by studying an autoassociative Potts network, as a model of memory storage distributed over several cortical areas, and testing whether the network can represent both units of word and signs, reflecting the types and patterns of errors made by individuals with POB impairment.

Non-hexagonal neural dynamics in vowel space.

Are the grid cells discovered in rodents relevant to human cognition? Following up on two seminal studies by others, we aimed to check whether an approximate 6-fold, grid-like symmetry shows up in the cortical activity of humans who "navigate" between vowels, given that vowel space can be approximated with a continuous trapezoidal 2D manifold, spanned by the first and second formant frequencies. We created 30 vowel trajectories in the assumedly flat central portion of the trapezoid. Each of these trajectories had a duration of 240 milliseconds, with a steady start and end point on the perimeter of a "wheel". We hypothesized that if the neural representation of this "box" is similar to that of rodent grid units, there should be an at least partial hexagonal (6-fold) symmetry in the EEG response of participants who navigate it. We have not found any dominant n-fold symmetry, however, but instead, using PCAs, we find indications that the vowel representation may reflect phonetic features, as positioned on the vowel manifold. The suggestion, therefore, is that vowels are encoded in relation to their salient sensory-perceptual variables, and are not assigned to arbitrary grid-like abstract maps. Finally, we explored the relationship between the first PCA eigenvector and putative vowel attractors for native Italian speakers, who served as the subjects in our study.

Partial coherence and frustration in self-organizing spherical grids.

Nearby grid cells have been observed to express a remarkable degree of long-range order, which is often idealized as extending potentially to infinity. Yet their strict periodic firing and ensemble coherence are theoretically possible only in flat environments, much unlike the burrows which rodents usually live in. Are the symmetrical, coherent grid maps inferred in the lab relevant to chart their way in their natural habitat? We consider spheres as simple models of curved environments and waiting for the appropriate experiments to be performed, we use our adaptation model to predict what grid maps would emerge in a network with the same type of recurrent connections, which on the plane produce coherence among the units. We find that on the sphere such connections distort the maps that single grid units would express on their own, and aggregate them into clusters. When remapping to a different spherical environment, units in each cluster maintain only partial coherence, similar to what is observed in disordered materials, such as spin glasses.

Can Grid Cell Ensembles Represent Multiple Spaces?

The way grid cells represent space in the rodent brain has been a striking discovery, with theoretical implications still unclear. Unlike hippocampal place cells, which are known to encode multiple, environment-dependent spatial maps, grid cells have been widely believed to encode space through a single low-dimensional manifold, in which coactivity relations between different neurons are preserved when the environment is changed. Does it have to be so? Here, we compute, using two alternative mathematical models, the storage capacity of a population of grid-like units, embedded in a continuous attractor neural network, for multiple spatial maps. We show that distinct representations of multiple environments can coexist, as existing models for grid cells have the potential to express several sets of hexagonal grid patterns, challenging the view of a universal grid map. This suggests that a population of grid cells can encode multiple noncongruent metric relationships, a feature that could in principle allow a grid-like code to represent environments with a variety of different geometries and possibly conceptual and cognitive spaces, which may be expected to entail such context-dependent metric relationships.