Redundancy and blocking in the spatial domain: a connectionist model

How can the observations of spatial blocking (Rodrigo, Chamizo, McLaren & Mackintosh, 1997) and cue redundancy (O’Keefe and Conway, 1978) be reconciled within the framework provided by an error-correcting, connectionist account of spatial navigation? I show that an implementation of McLaren’s (1995) better beta model can serve this purpose, and examine some of the implications for spatial learning and memory. In this paper I tackle an issue in spatial navigation. How can a system for navigation be devised so that it combines the ability to use cue combinations in a redundant fashion (O’Keefe and Conway, 1978) at the same time as allowing for the spatial equivalent of blocking (Rodrigo et al, 1997). I start by briefly describing both sets of results before offering a connectionist implementation of the approach taken in McLaren (1995) as a partial solution to the problem. O’Keefe and Conway’s (1978) demonstration of redundancy in spatial navigation is illustrated in Figure 1. O’Keefe and Conway were able to show that there were some hippocampal cells (units), that they termed place cells, that fired when the animal was in a specific location on the maze used in training and testing. These regions were defined by the four landmarks placed around the maze, and O’Keefe and Conway made sure that this was the case by rotating the maze relative to extra-maze cues from time to time. The key result was that any two of these landmarks were sufficient for at least some of the place cells to fire when the animal entered the appropriate region. This is a demonstration of redundancy because the coding of spatial location is not critically dependent on any given cue, either singly or in combination with other cues. In this case any permutation of two cues from four was sufficient to enable navigation, making the system serving as the basis for navigation robust to cue removal or alteration. Rodrigo, Chamizo, McLaren, & Mackintosh (1997) were the first to demonstrate the spatial equivalent of blocking. The apparatus used is illustrated in Figure 2. This shows a swimming pool in a rectangular room, with four potential landmarks placed around the pool in the positions shown.