Three Dimensional Frames of References Transformations using Recurrent Populations of Neurons

This work investigates whether population vector coding, a distributed computational paradigm, could be a principle mechanism for performing sensorimotor and frames of reference transformations. This paper presents a multilayer neural network that can perform arbitrary three-dimensional rotations and translations. We demonstrate, both formally and numerically, that the non-linearity of these transformations can be resolved thanks to the recurrent and concurrent activities of continuous populations of neurons.