Encoding Longer-term Contextual Sensorimotor Information in a Predictive Coding Model

Abstract

Studies suggest that within the hierarchical architecture, the topological higher level possibly represents a conscious category of the current sensory events with a slower changing activities. They attempt to predict the activities on the lower level by relaying the predicted information. On the other hand, the incoming sensory information corrects such prediction of the events on the higher level by the novel or surprising signal. We propose a predictive hierarchical artificial neural network model that examines this hypothesis on neurorobotic platforms, based on the AFA-PredNet model. In this neural network model, there are different temporal scales of predictions exist on different levels of the hierarchical predictive coding, which are defined in the temporal parameters in the neurons. Also, both the fastand the slow-changing neural activities are modulated by the active motor activities. A neurorobotic experiment based on the architecture was also conducted based on the data collected from the VRep simulator.

Disentanglement in conceptual space during sensorimotor interaction

“…At the first experiments, we will examine the predictive ability of the VAE within the predictive framework of the AFA‐PredNet. Similarly, as we did in AFA‐PredNet [23] and MT‐AFA‐PredNet [24, 25], the first results from the generative models to be examined are the generated images. The training was done using Adam optimisation [28] and other hyperparameters were defined as the table follows (Tables 1 and 2).…”

“…The proposed variational action feedback augmented predictive network (VAFA‐PredNet) is a hierarchical architecture. It is an extension of the previous versions of AFA‐PredNet [23] and multiple time‐scale action feedback augmented predictive network (MT‐AFA)‐PredNet [24, 25]. Inspired by the predictive coding framework, all of these hierarchical neural networks integrate the perception and action in the predictive common‐coding framework [22]: the predicted perception is part of the results of the internal visual memory and the voluntary motor actions.…”

“…Modelling the likelihood representation on top of the generative units, the VAE extracts the representation of the memory of the topmost convLSTM unit. Since in the MT‐AFA‐PredNet [24, 25], the slower‐changing neural updates on the topmost of the convLSTM unit only adapts very small differences, whereas the time elapse, the input and reconstruct images in the VAE model can exist very small differences. The detailed introduction of the VAE unit please is shown in Section 2.2.…”