Inference and segmentation in cortical processing

Abstract

We present a modelling framework for cortical processing aimed at understanding how, maintaining biological plausibility, neural network models can: (a) approximate general inference algorithms like belief propagation, combining bottom-up and top-down information, (b) solve Rosenblatt's classical superposition problem, which we link to the binding problem, and (c) do so based on an unsupervised learning approach. The framework leads to two related models: the first model shows that the use of top-down feedback significantly improves the network's ability to perform inference of corrupted inputs; the second model, including oscillatory behavior in the processing units, shows that the superposition problem can be efficiently solved based on the unit's phases.

Unsupervised Segmentation With Dynamical Units

“…Moreover, the scheme presented here lends itself relatively easily to generalizations, in particular, by extending the feedback connections to affect not only the phase of lower units, but also their amplitude. We are indeed working towards an integrated model to account for both segmentation and inference in the presence of partial information [30]. We have also shown the ability of the network presented in this paper to achieve translation invariant encoding of objects, with a minor modification to the learning rule [27].…”

“…There are several fruitful extensions of our model, which we have begun pursuing [27], [30]. Further work needs to be done to investigate the robustness of oscillatory networks in the handling of geometric distortions.…”

“…Another possibility is to use amplitude information in the feedback connections. An early investigation in this direction is presented in [30].…”

“…A promising direction to explore is the incorporation of amplitude in the feedback connections, as explored in [30]. In addition, it may be necessary to increase the number of layers in the network from two to three or more to effectively utilize feedback information.…”

An optimization approach to achieve unsupervised segmentation and binding in a dynamical network

“…Moreover, the scheme presented here lends itself relatively easily to generalizations, in particular by extending the feedback connections to affect not only the phase of lower units, but also their amplitude. We are indeed working towards an integrated model to account for both segmentation and inference in the presence of partial information [17]. We have also shown the ability of the network presented in this paper to achieve translation invariant encoding of objects, with a minor modification to the learning rule [18].…”