The authors draw together the results of a series of detailed computational studies and show how they are contributing to the development of a theory of hippocampal function. A new part of the theory introduced here is a quantitative analysis of how backprojections from the hippocampus to the neocortex could lead to the recall of recent memories. The theory is then compared with other theories of hippocampal function. First, what is computed by the hippocampus is considered. The hypothesis the authors advocate, on the basis of the effects of damage to the hippocampus and neuronal activity recorded in it, is that it is involved in the formation of new memories by acting as an intermediate-term buffer store for information about episodes, particularly for spatial, but probably also for some nonspatial, information. The authors analyze how the hippocampus could perform this function, by producing a computational theory of how it operates, based on neuroanatomical and neurophysiological information about the different neuronal systems con-tained within the hippocampus. Key hypotheses are that the CA3 pyramidal cells operate as a single autoassociation network to store new episodic information as it arrives via a number of specialized preprocessing stages from many association areas of the cerebral cortex, and that the dentate

. Much evidence indicates that recognition memory involves two separable processes, recollection and familiarity discrimination, with familiarity discrimination being dependent on the perirhinal cortex of the temporal lobe. Here, we describe a new neural network model designed to mimic the response patterns of perirhinal neurons that signal information concerning the novelty or familiarity of stimuli. The model achieves very fast and accurate familiarity discrimination while employing biologically plausible parameters and Hebbian learning rules. The fact that the activity patterns of the model's simulated neurons are closely similar to those of neurons recorded from the primate perirhinal cortex indicates that this brain region could discriminate familiarity using principles akin to those of the model. If so, the capacity of the model establishes that the perirhinal cortex alone may discriminate the familiarity of many more stimuli than current neural network models indicate could be recalled (recollected) by all the remaining areas of the cerebral cortex. This efficiency and speed of detecting novelty provides an evolutionary advantage, thereby providing a reason for the existence of a familiarity discrimination network in addition to networks used for recollection. Keywords: recognition memory, novelty detection, hippocampal region, computational model, spike-response model 1.

There has been a controversy on whether the standard neuro-spiking models are consistent with the irregular spiking of cortical neurons. In a previous study, we proposed examining this consistency on the basis of the high order statistics of the inter-spike intervals, as represented by the coe#cient of variation and the skewness coe#cient. In that study we found that a leaky integrate-and-fire model incorporating the assumption of temporally uncorrelated inputs is not able to account for the spiking data recorded from a monkey prefrontal cortex. In the present paper, we attempt to revise the neuro-spiking model so as to make it consistent with the biological data. Here we consider the correlation coe#cient of consecutive inter-spike intervals, which was ignored in previous studies. Considering three statistical coe#cients, we conclude that the leaky integrate-and-fire model with temporally correlated inputs does account for the biological data. The correlation time scale of the inputs ...

This study compares the efficiency and plausibility of published computational models of familiarity discrimination in the perirhinal cortex. Substantial evidence indicates that the perirhinal cortex is involved in both the familiarity discrimination aspect of recognition memory and in perceptual functions involved with representations of complete stimuli (i.e., object identification). Published models of how the perirhinal cortex may perform familiarity discrimination can be divided into two groups. The first group assumes that a proportion of perirhinal neurons form a network specialised just for familiarity discrimination (these models may be based on Hebbian or anti-Hebbian synaptic plasticity). In contrast, the second group assumes that both familiarity discrimination and learning representations of complete stimuli are performed within a single combined network. This study establishes that when the responses of neurons that provide input to the familiarity discrimination network are correlated (as indicated by experimental data), specialised networks based on anti-Hebbian learning may recognise the previous occurrence of many more stimuli (i.e., have a capacity up to thousands of times larger) than specialised networks based on Hebbian learning. The currently published combined models do not learn an optimal stimulus representation (they do not fully extract statistically independent features), and hence their capacities are even lower than those of the specialised models based on Hebbian learning. Hence, the combined models published thus far are critically less efficient than the specialised models based on anti-Hebbian learning. This study also compares the consistency of the models with experimental observations concerning what is kno...

We investigate the properties of an autoassociative network of thresholdlinear units whose synaptic connectivity is spatially structured and asymmetric. Since the methods of equilibrium statistical mechanics cannot be applied to such a network due to the lack of a Hamiltonian, we approach the problem through a signal-to-noise analysis, that we adapt to spatially organized networks. The conditions are analyzed for the appearance of stable, spatially non-uniform profiles of activity with large overlaps with one of the stored patterns. It is also shown, with simulations and analytic results, that the storage capacity does not decrease much when the connectivity of the network becomes short range. In addition, the method used here enables us to calculate exactly the storage capacity of a randomly connected network with arbitrary degree of dilution. 1

First, the information represented in the primate hippocampus, and what is computed by the primate hippocampus, are considered.

A fundamental problem in neuroscience is understanding how working memory—the ability to store information at intermediate timescales, like tens of seconds—is implemented in realistic neuronal networks. The most likely candidate mechanism is the attractor network, and a great deal of effort has gone toward investigating it theoretically. Yet, despite almost a quarter century of intense work, attractor networks are not fully understood. In particular, there are still two unanswered questions. First, how is it that attractor networks exhibit irregular firing, as is observed experimentally during working memory tasks? And second, how many memories can be stored under biologically realistic conditions? Here we answer both questions by studying an attractor neural network in which inhibition and excitation balance each other. Using mean-field analysis, we derive a three-variable description of attractor networks. From this description it follows that irregular firing can exist only if the number of neurons involved in a memory is large. The same mean-field analysis also shows that the number of memories that can be stored in a network scales with the number of excitatory connections, a result that has been suggested for simple models but never shown for realistic ones. Both of these predictions are verified using simulations with large networks of spiking neurons.

At the transition from early reptilian ancestors to primordial mammals, the areas of sensory cortex that process topographic modalities acquire the laminar structure of isocortex. A prominent step in lamination is granulation, whereby the formerly unique principal layer of pyramidal cells is split by the insertion of a new layer of excitatory, but intrinsic, granule cells, layer IV. I consider the hypothesis that granulation, and the differentiation between supra- and infra-granular pyramidal layers, may be advantageous to support fine topography in their sensory maps. Fine topography implies a generic distinction between "where" information, explicitly mapped on the cortical sheet, and "what" information, represented in a distributed fashion as a distinct firing pattern across neurons. These patterns can be stored on recurrent collaterals in the cortex, and such memory can help substantially in the analysis of current sensory input. The simulation of a simplified network model demonstrates that a non-laminated patch of cortex must compromise between transmitting "where" information or retrieving "what" information. The simulation of a modified model including differentiation of a granular layer shows a modest but significant quantitative advantage, expressed as a less severe trade-off between "what" and "where". The further connectivity differentiation between infra-granular and supra-granular pyramidal layers is shown to match the mix of "what" and "where" information optimal for their respective target structures.

We start from Landauer's realization that "information is physical", i.e. that computation cannot be disentangled from the physical system used to perform it, and ask what the capabilities of physical systems really are. In particular, is it possible to design a physical system which is able to solve hard (i.e. NP-complete) problems more e#ciently than conventional computers? Chaotic physical systems (such as the weather) are hard to predict or simulate, but we find that they are also hard to control. The requirement of control turns out to pin down the non-conventional options to either neural networks or quantum computers. Alternatively, we can give up the possibility of control in favour of a system which is basically chaotic, but is able to settle at a solution if it reaches one. However, systems of this type appear inevitably to perform a type of stochastic local search.