Biology Reference
In-Depth Information
With its extensive recurrent collateral connectivity the CA3 region is suggested to act as
an autoassociation memory which enables episodic memories to be formed and stored in
hippocampus (Marr, 1971; McNaughton & Morris, 1987; Rolls, 1991; Treves & Rolls, 1992).
The memory for sequences is then determined by the synaptic modifications in the recurrent
collateral synapses (Rolls & Treves, 1998; Rolls & Deco, 2002). Subsequently the recurrent
collateral activity allows for the retrieval of a whole representation to be initiated by the
activation of some small part of the same representation. This property, known as “pattern
completion”, in principle allows a memory to be retrieved in full when the animal is presented
with only some reminders of it. Therefore the Hebbian learning mechanisms occurring in the
recurrent connections allow the full retrieval of a representation based on only a partial,
fragmented input. An important property of the autoassociation model of the CA3 recurrent
collateral network is that the retrieval can be symmetric - the whole memory sequence can be
retrieved from any part. For example, in an object-place autoassociation memory, an object
could be recalled from a place retrieval cue, and
vice versa
. As the hippocampus operates
effectively as a single network, it can allow arbitrary associations between inputs originating
from very different parts of the cerebral cortex to be formed. These might involve
associations between information originating in the temporal visual cortex about the presence
of an object, and information originating in the parietal cortex about where it is; hence
hippocampus enables different memories to be stored in a certain sequence (Rolls & Kesner,
2006).
Figure 6. Neural network architecture for two-dimensional continuous attractor models of place cells. A
recurrent network of place cells with firing rates r(p) receives external inputs from three sources: (1)
visual system - I(v), (2) population of head direction cells with firing rates r(hd), and (3) population of
forward velocity cells with firing rates r(fv). The recurrent weights between the place cells are denoted
by w(rc), and the idiothetic weights to the place cells from the forward velocity cells and head direction
cells are denoted by w(fv) (adapted from (Stringer
et al.
, 2004) and (Rolls & Kesner, 2006)).
Autoassociation models of the CA3 recurrent collateral network also implement the
ability to maintain the firing of neurons using excitatory recurrent collateral connections. A
stable attractor can maintain one memory active in this way for a considerable period, until a
new input pushes the attractor to represent a new location or memory (Treves & Rolls, 1991, ,
1992). There is evidence implicating the hippocampus in mediating associations across time
