================================================================================================== Alexei V. Samsonovich1, and Giorgio A. Ascoli1 (2005) " A simple neural network model of the hippocampus suggesting its pathfinding role in episodic memory retrieval." Learning Memory No. 12 Published online, Cold Spring Harbor Laboratory Press. pp. 193-208. ================================================================================================== [abstract] "We propose a mechanism of reconstruction of the context of experience involving a search for a nearly shortest path in the space of remembered contexts." "It is demonstrated how a nearly shortest path is quickly found in a familiar environment." [Introduction] "Hippocampal pyramidal cells, called place cells in rodents, each selectively fire at a high rate when the animal is located in a particular spatial domain, called a place field of the place cell. This type of firing is observed during active maze running (O'Keefe and Dostrovsky 1971), and to some extent, during slow-wave sleep states (Wilson and Mc-Naughton 1994; Jarosiewicz et al. 2002; Lee and Wilson 2002; Jarosiewicz and Skaggs 2004). Equivalent or similar phenomena were observed in primates (Rolls and OfMara 1995; Robertson et al. 1998), including humans (Ekstrom et al. 2003). The system of place fields forms a cognitive map of an environment (O'Keefe and Nadel 1978, 1979)." "Our assumption is that an episodic memory trace may not be retrievable without reinstatement of a missing contextual key -- a key that is not associated with the pointer. We propose a theory explaining how this key is found by the hippocampus, why finding it may be impossible without the hippocampus, and how this relates to spatial navigation. "This information needs to be acquired from a set of other episodic memories that serve as a key for understanding the target episode. This acquisition can be subserved by the system taking a quick path through the key set, briefly reactivating necessary memories one by one. Finding a correct path may be crucial for a successful retrieval. How can one decide which path is "correct" -- In this study, we say that there is a connection from episodic memory A to episodic memory B, if, given an explicit awareness and an understanding of A, the system is capable of immediate retrieval and understanding of B without consulting any additional episodic knowledge (in contrast, available semantic knowledge may be used). Then, a rule that ensures the retrieval of an episodic memory X, given an initial state of awareness Y, is to follow a connected path from Y to X, i.e., a continuous path composed of relevant remembered episodes and connections among them." [Sharp waves and associative spatial learning] "Experimental data suggest that time-compressed replay of the recent trajectory may occur during sharp waves in sleep and wake states" [Phase precession: Exploration of feasible actions?] Many theories have been recently proposed regarding the mechanisms and the function of PP, including its role in learning associations among sequentially visited places (Skaggs and Mc-Naughton 1996) and its possible origin from the modulation of the sensory input (Burgess et al. 1994; Mehta et al. 2002). Several interesting features of PP have been established experimentally, including the following (Skaggs et al. 1996). PP is not sensitive to visual or other sensory input, e.g., it is observed in complete darkness. PP is observed during linear one-dimensional unidirectional motion as well as during random two-dimensional foraging. In both cases, its direction is determined by the direction of the head of the animal, but the spatial amplitude of onedimensional PP is approximately twice as large as that of twodimensional PP (Fig. 1D,E). -------------------------------------------------------------------------------------------------- McNaughton, B. L., F. B. Battaglia, O. Jensen, E. I. Moser, and M. Moser. 2006. Path integration and the neural basis of the ecognitive map.f Nat. Rev. Neurosci. 7: 663.678. -------------------------------------------------------------------------------------------------- According to a common description of this phenomenon (Skaggs et al. 1996) Skaggs, W.E. and McNaughton, B.L. 1996. Replay of neuronal firing sequences in rat hippocampus during sleep following spatial experience. Science 271: 1870.1873. => not available -------------------------------------------------------------------------------------------------- An alternative, equivalent description of the same phenomenon can be given in terms of a population code (Samsonovich and McNaughton 1997). Samsonovich, A. and McNaughton, B.L. 1997. Path integration and cognitive mapping in a continuous attractor neural network model. J. Neurosci. 17: 5900.5920. => Done [Materials and Methods] [Fig. 2 caption] Sensory module stimulates the CA3 unit S that represents the current location (or the currently perceived context) on the cognitive map. [Model principles, organization, and dynamics] -------------------------------------------------------------------------------------------------- We start by defining a simple model of spatial learning and navigation. The central component of the model is the hippocampal network In the application of the model to spatial navigation, >>each<< small neighborhood in the environment has a unique representation in the system being associated with a particular CA3 unit and a particular CA1 unit. This is consistent with the (now widely accepted) view of rodent hippocampal pyramidal cells as place cells. Place cells have preferred locations in an environment characterized by a high firing rate of the cell. We interpret the gradient of a firing-rate distribution of each place cell as encoded gdriving directionsh to the point of a maximal firing rate for this cell, which can be an arbitrarily selected point of the environment. In order to find a way to a goal, the system glistensh to the background activity of goal-related cells, while exploring alternative feasible actions during sequential theta cycles. wikipedia -------------------------------------------------------------------------------------------------- Another school, led by John O'Keefe, have suggested that theta is part of the mechanism animals use to keep track of their location within the environment. -------------------------------------------------------------------------------------------------- At this point, the activity mode changes from regular, theta-modulated place-cell firing to LIA with gsharp waveh reactivation of recently active CA3 place cells, with a firing rate proportional to the recency of their last strong activation. In addition, CA1 cells representing the current location are also reactivated (Buzsaki 1989). As a result, CA3 place cells become associated with the selected CA1 cell whose place field represents a potential future goal. The strength of the associations in this model is proportional to the recency of a place-cell firing during exploration: Thus, if the exploration phase continues until a preselected goal location is reached, the CA3 model units acquire potentiated, weighted connections to the CA1 representation of the goal. At this point, the network of place cells actually provides a capacity for finding a short path to this goal location, starting from any given location within the environment. This is accomplished by a simple algorithm; at each location, several randomly generated possible local moves are explored via PP, and the move that produces the strongest excitation of the place cell associated with the goal is selected and performed. This process is repeated until the goal is reached. In the case of two-dimensional spatial navigation, the PP direction fluctuates within a standard set (gright,h gleft,h etc.) For the sake of parsimony, CA3-to-CA3 synapses are only included in the model of memory retrieval, and not in the model of spatial navigation.