Emulating Nature: Models of Hippocampus
This chapter covers the state of the art models of the rodent hippocampus and their ability to solve each component of the mapping and navigation problem. Rodent hippocampal models can be separated into two groups. The first contains models that have been
- PDF / 2,144,568 Bytes
- 13 Pages / 430 x 660 pts Page_size
- 88 Downloads / 163 Views
This chapter covers the state of the art models of the rodent hippocampus and their ability to solve each component of the mapping and navigation problem. Rodent hippocampal models can be separated into two groups. The first contains models that have been developed using theories of the neuronal mechanisms by which rats navigate and create a map. By testing these models in simulation or on a robot these theories can be validated or disproved. These models can also be used to make predictions that can be tested by further rodent experiments. Most of the models described in this chapter fall into this group. The second group of models is much smaller and contains models that have been developed for practical purposes, such as for controlling a mobile robot. These models vary in their trade-off between biological faithfulness and practical performance, with none matching the mapping performance of the top probabilistic techniques. This chapter also describes these models, with Chapter 6 discussing the implications of their limited practical performance.
5.1 Head Direction and Place Cells – State of the Art Models of the rodent hippocampus often consist of head direction and place cell structures. Because head direction cells relate to the rodent head’s one-dimensional orientation state, the computational algorithms required to simulate them are generally less complex than those required to simulate place cells, which relate to the rat’s twodimensional location state. 5.1.1 Attractor Networks Models of head direction and place cells often use some form of attractor network (Redish, Elga et al. 1996; Zhang 1996; Samsonovich and McNaughton 1997; Stringer, Rolls et al. 2002; Stringer, Trappenberg et al. 2002). Typically an array of cells is used, with nearby cells linked by strong excitatory connections, and distant cells linked by inhibitory connections. The stable state of such a network consists of a single cluster of active cells in a shaped peak distribution. The orientation represented by the current activity state can be determined in several ways, such as through population vector decoding or by simply picking the most highly activated cell. M.J. Milford: Robot Navigation from Nature, STAR 41, pp. 41–53, 2008. © Springer-Verlag Berlin Heidelberg 2008 springerlink.com
42
Emulating Nature: Models of Hippocampus
Fig. 5.1. Attractor network model of the rodent head direction system. Activity in the outer ring of head direction cells encodes orientation. The inner two rings are vestibular cells that respond to angular velocities. Links between visual cells and head direction cells are modified using a Hebbian learning rule. Head direction cells also have strong intrinsic connections that stabilise the system to a single localised cluster of active cells (Skaggs, Knierim et al. 1995). Reprinted with permission from The MIT Press.
5.1.2 Path Integration Activity injected into an attractor network near either side of the current activity peak will cause the peak to shift towards the injected activity through the net
Data Loading...
