Neural mechanism of dynamic responses of neurons in inferior temporal cortex in face perception

Abstract

Understanding the neural mechanisms of object and face recognition is one of the fundamental challenges of visual neuroscience. The neurons in inferior temporal (IT) cortex have been reported to exhibit dynamic responses to face stimuli. However, little is known about how the dynamic properties of IT neurons emerge in the face information processing. To address this issue, we made a model of IT cortex, which performs face perception via an interaction between different IT networks. The model was based on the face information processed by three resolution maps in early visual areas. The network model of IT cortex consists of four kinds of networks, in which the information about a whole face is combined with the information about its face parts and their arrangements. We show here that the learning of face stimuli makes the functional connections between these IT networks, causing a high spike correlation of IT neuron pairs. A dynamic property of subthreshold membrane potential of IT neuron, produced by Hodgkin–Huxley model, enables the coordination of temporal information without changing the firing rate, providing the basis of the mechanism underlying face perception. We show also that the hierarchical processing of face information allows IT cortex to perform a “coarse-to-fine” processing of face information. The results presented here seem to be compatible with experimental data about dynamic properties of IT neurons.

Appendix: The synaptic weights between neurons in V4X and ITX (X = B, M, F)

The synaptic weights between neurons in V4X and ITX (X = B, M, F) were determined by the learning rule of Kohonen’s self-organized map (Kohonen 1995). The V4B has a retinotopic map consisting of N_V4B x N_V4B neurons, in which each neuron has the output of 0 or 1, depending on whether it receives the input elicited by a face stimulus, as shown in Fig. 2a. The synaptic weight of the connection from (i, j)th neuron in V4B to the neuron within kth uint in ITB layer, $w_{V4B - ITB} (k,ij)$, is updated by $\Updelta w_{V4B - ITB} (k,ij)$, given by

$$ \Updelta w_{V4B - ITB} (k,ij) = \lambda_{1} (X_{ij,V4B} - w_{V4B - ITB} (k,ij))\,for\,k = k*, $$

(11)

$$ k^{*} = \arg \,\mathop {\min }\limits_{k} \,d_{k} $$

(12)

$$ d_{k} = \sqrt {\sum\limits_{ij} {(X_{ij,V4B} - w_{V4B - ITB} (k,ij))^{2} } } , $$

(13)

where X_ij,V4B is the output of (i, j)th V4B neuron and λ₁ is the learning rate. The parameter values used were: N_V4B = 11, λ₁ = 0.01.

The V4M has a retinotopic map consisting of N_V4M x N_V4M neurons, which has the same size as V4B, as shown in Fig. 2b. The V4M neurons encode the information about the relative position of the face parts chosen by attention. When attention is directed to the position (i₀, j₀) included in a face part, the output of (i, j)th V4M neuron, $X_{ij,V4M}$, is determined by

$$ X_{ij,V4M} = \left\{ {\begin{array}{*{20}c} {1,} & {\left| {i - i_{0} } \right| \le d,\,or\,\left| {j - j_{0} } \right| \le d} \\ {0,} & {otherwise.} \\ \end{array} } \right. $$

(14)

Then the synaptic weight of the connection from (i, j)th neuron in V4M to kth neuron within lth unit in ITM layer is updated by $\Updelta w_{V4M - ITM} (kl,ij)$, given by

$$ \Updelta w_{V4M - ITM} (kl,ij) = \lambda_{2} (X_{ij,V4M} - w_{V4M - ITM} (kl,ij))\,for\,(k,.l) = (k*,l*), $$

(15)

$$ (k*,\,l*) = \arg \,\mathop {\min }\limits_{kl} \,d_{kl} , $$

(16)

$$ d_{kl} = \sqrt {\sum\limits_{ij} {(X_{ij,V4M} - w_{V4M - ITM} (kl,ij))^{2} } } . $$

(17)

The synaptic weights of ITM neurons within unit were slightly modulated so as to have different sensitivities to the output of V4M. The parameter values were: N_V4M = 11, d = 1, and λ₂ = 0.01.

The V4F consists of N_V4F x N_V4F hypercolumns, each of which contains N_L columns, as shown in Fig. 2c. These neurons in the columns encode local curvatures of a face part chosen by attention. The V4F represents only local features, in contrast to the representations of global face by V4B and V4M. Thus the local face features encoded by V4F is a complement to the face features encoded by V4B and V4M. The synaptic weights between ITF and V4F were determined by the learning rules similar to Eqs. (15–17). The parameter values used were: N_V4F = 3, N_L = 10, and the learning rate was set at 0.01.