|
Résumé :
If the standard model for long term memory claims that we need two memory storage systems, one for episodic memory, the other for semantic memory, one must admit that this is still an hypothesis even if almost every work on memory is based on this assumption. So, we tried to evaluate if a back propagation network, which is able to generalize for learning the world structure, can also have a rapid and specific item memory (i.e. episodic memory). We hypothesized that a crucial factor in this evaluation is to take into account that for human learning occurs in a non arbitrary world, thus we compared the item memory capacity with structured and non structured data. In the actual state, the results of this study are mixed regarding the back propagation network capabilities as a system able to generalize and able to learn quickly with a specific item memory. So we need in the future to investigate the network behaviour with probabilistic structure, in order to further determine if a single memory storage can indeed account for both rapid memorization of exemplars and human generalization capacities.
|
|
Mots clés :
Memory model, structuration effect, backpropagation network, episodic memory, generalization
|