In the inferior temporal cortex of the monkey, ‘pair-recall’ neurons which exhibit prospective activity for the target during the delay period of a pair-association task have been found. To explain this and other physiological findings, we previously constructed a model of pair-association memory consisting of two interactive networks. The present paper reports that recent empirical data on the time course of the pair-recall activity accord very well with the prediction of our model. This strongly suggests that memories are embedded in trajectory attractors in area TE, implying that the learning signal necessary for forming them is sent backward from the perirhinal cortex.
Kernel machines such as support vector machines require a kernel function defined a priori. For successful learning, the kernel function should be properly designed according to prior knowledge. This paper gives an extensive review on recently proposed kernel design strategies such as convolution kernels, marginalized kernels, Fisher kernels, TOP kernels, string kernels and diffusion kernels.