this paper with describing linked temporal sequences of events.) The hippocampus is also implicated in spatial memory. For example, damage to the hippocampal system in monkeys produces de# cits in learning about where objects are and where responses must be made (Rolls 1996; Gaffan 1998), and in rats produces spatial learning de# cits (Martin et al. 2000). Neurophysiologically, hippocampal neurons in rats respond to the place where the animal is located (O'Keefe 1990; Kubie & Muller 1991; Wilson & McNaughton 1993), and in primates to the place being viewed (Rolls et al. 1997; Rolls 1999). It has thus been a long-standing question about whether the hippocampus and nearby temporal lobe structures are involved in episodic memory or spatial function. In this paper we show that this question can be resolved by revealing that a single neural network can implement both episodic and spatial memory

Abstract not found

This paper presents a method for a mobile robot to construct and localize relative to a "cognitive map", where the cognitive map is assumed to be a representational structure that encodes both spatial and behavioral information. The localization is performed by applying a generic Bayes filter. The cognitive map was implemented within a behavior-based robotic system, providing a new behavior that allows the robot to anticipate future events using the cognitive map. One of the prominent advantages of this approach is elimination of the pose sensor usage (e.g., shaft encoder, compass, GPS, etc.), which is known for its limitations and proneness to various errors. A preliminary experiment was conducted in simulation and its promising results are discussed.

A computational model of the hippocampal function in spatial learning is presented. A spatial representation is incrementally acquired during exploration. Visual and self-motion information is fed into a network of rate-coded neurons. A consistent and stable place code emerges by unsupervised Hebbian learning between place- and head direction cells. Based on this representation, goal-oriented navigation is learnt by applying a reward-based learning mechanism between the hippocampus and nucleus accumbens. The model, validated on a real and simulated robot, successfully localises itself by recalibrating its path integrator using visual input. A navigation map is learnt after about 20 trials, comparable to rats in the water maze. In contrast to previous works, this system processes realistic visual input. No compass is needed for localisation and the reward-based learning mechanism extends discrete navigation models to continuous space. The model reproduces experimental findings and suggests several neurophysiological and behavioural predictions in the rat. 1

The place fields of hippocampal cells in old animals sometimes change when an animal is removed from and then returned to an environment [ Barnes et al., 1997 ] . The ensemble correlation between two sequential visits to the same environment shows a strong bimodality for old animals (near 0, indicative of remapping, and greater than 0.7, indicative of a similar representation between experiences), but a strong unimodality for young animals (greater than 0.7, indicative of a similar representation between experiences). One explanation for this is the multi-map hypothesis in which multiple maps are encoded in the hippocampus: old animals may sometimes be returning to the wrong map. A theory proposed by Samsonovich and McNaughton (1997) suggests that the Barnes et al. experiment implies that the maps are pre-wired in the CA3 region of hippocampus. Here, we offer an alternative explanation in which orthogonalization properties in the dentate gyrus (DG) region of hippocampus interact with e...

It is well-known that identical sensory input under dierent perceptual, behavioral or contextual conditions can produce distinct patterns of activity in the place cells of the rodent hippocampus. However, the mechanisms underlying this have not been completely claried. A recent experiment has shown that place cell activity on a 3-arm maze exhibits hysteresis as the maze is rotated with respect to distal cues. The apparent angular extent of a place eld is greater when a maze arm rotates out of it than when it rotates back into the eld. In this report, we present a simple attractor-based model of the hippocampus that reproduces this hysteresis phenomenon. The model allows us to make predictions about changes in the hysteresis eect as the animal becomes more familiar with the maze in several orientations. It also has implications for the place eld remapping phenomenon seen in many hippocampal experiments. Keywords: Hippocampus, place cells, spatial representation, attractor network...

Cells in several areas of the hippocampal formation show place specific firing patterns, and are thought to form a distributed representation of an animal’s current location in an environment. Experimental results suggest that this representation is continually updated even in complete darkness, indicating the presence of a path integration mechanism in the rat. Adopting the Neural Engineering Framework (NEF) presented by Eliasmith and Anderson (2003) we derive a novel attractor network model of path integration, using heterogeneous spiking neurons. The network we derive incorporates representation and updating of position into a single layer of neurons, eliminating the need for a large external control population, and without making use of multiplicative synapses. An efficient and biologically plausible control mechanism results directly from applying the principles of the NEF. We simulate the network for a variety of inputs, analyze its performance, and give three testable predictions of our model.

Attractor networks, which map an input space to a discrete output space, are useful for pattern completion---cleaning up noisy or missing input features. However, designing a net to have a given set of attractors is notoriously tricky; training procedures are CPU intensive and often produce spurious attractors and ill-conditioned attractor basins. These difficulties occur because each connection in the network participates in the encoding of multiple attractors. We describe an alternative formulation of attractor networks in which the encoding of knowledge is local, not distributed. Although localist attractor networks have similar dynamics to their distributed counterparts, they are much easier to work with and interpret. We propose a statistical formulation of localist attractor net dynamics, which yields a convergence proof and a mathematical interpretation of model parameters. We present simulation experiments that explore the behavior of localist attractor networks, show...

Previous research has identified a variety of strategies used by novice and experienced navigators in making cardinal direction judgments (Gugerty, Brooks, & Treadaway, 2004). We developed an ACT-R cognitive model of some of these strategies that instantiated a number of concepts from research in spatial cognition, including a visual-short-term-memory buffer overlaid on a perceptual buffer, an egocentric reference frame in visual-short-term-memory, storage of categorical spatial information in visual-short-term-memory, and rotation of a mental compass in visual-short-termmemory. Response times predicted by the model fit well with the data of two groups, college students (N D 20) trained and practiced in the modeled strategies, and jet pilots (N D 4) with no strategy training. Thus, the cognitive model seems to provide an accurate description of important strategies for cardinal direction judgments. Additionally, it demonstrates how theoretical constructs in spatial cognition can be applied to a complex, realistic navigation task.

Abstract not found