08.02.2013 Views

Bernal S D_2010.pdf - University of Plymouth

Bernal S D_2010.pdf - University of Plymouth

Bernal S D_2010.pdf - University of Plymouth

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

3.4. EXISTING MODELS<br />

Section 3.4,2). Similariathc previous mapping, cortical columns implemenl the different nodes<br />

or variables {e.g. coding for a specific region <strong>of</strong> Ihe input image), while minicolumns represent<br />

the different possible states or features at that location {e.g. different orientations).<br />

According to the authors, projections from lower cortical to layer 4 levels are responsible for<br />

the storage and detection <strong>of</strong> coincidence patterns. The synaptic connections between these lay­<br />

ers represent the co-ocurrence <strong>of</strong> palicms on its inputs. Layer 4 then projects onto layer 2/3<br />

pyramidal ceUs which are assumed to behave as complex ceils which respond to invariant fea­<br />

tures or motion sequences. Thus, layer 2/3 is responsible for the calculation <strong>of</strong> the feedforward<br />

Markov chains' (groups nr sequences) states as suggested by the high density <strong>of</strong> lalerai con­<br />

nections. At the same time, anatomical connections suggest these neurons project the Markov<br />

chain information to higher cortical levels, and incorporate high-level information, received via<br />

layer 1 projections, into Ihc computation <strong>of</strong> the Markov chains. Layer 5 pyramidal neurons wiih<br />

dendriies in layers 1. 3 and 4 are responsible for the Belief calcuiaiion, hinally, layer 6 neurons<br />

with dendrites in layer 5 compute the feedback messages for hiwer regions.<br />

Further inspection <strong>of</strong> the proposed mappings reveals several key similarities and differences<br />

between them, which are summarized in Figure 3.14. Each variable or graph node Is roughly<br />

understood as a cortical functional column, containing smaller functional units or minicolumns,<br />

which correspond to the dlffcrenl variable states (George and Hawkins 2(K)9. Litvak and Ullman<br />

2009). Feedforward outgoing mes,sages from a node are assumed to originate from pyramidal<br />

cells in layer 2/3 (George and Hawkins 2009, Litvak and Ullman 2009, Lee and Mumford 2003,<br />

Friston et al. 2006). Feedback outgoing messages originate from pyramidal cells In the infra-<br />

granular layers (Frision cl al, 2006), either layer 5 (Lee and Mumford 2003) or layer 6 (George<br />

and Hawkins 2009). Feedforward incoming messages from lower cortical areas target layer 4<br />

neurons (Frision et al. 2006. Litvak and Uliman 2009, George and Hawkins 2009).<br />

However, ihere are two dlffereni neural popuhiions that could polenlially encode the incom­<br />

ing feedback messages from higher-levels; neurons in supragranular layers (Liivak and Ullman<br />

2009, Friston 2010), more precisely, in layer 1 according to George and Hawkins (2009); or<br />

neurons in infragranular layer 6 (Litvak and Ullman 2009, Friston et al. 2006). Both <strong>of</strong> them<br />

135

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!