Toward predictive machine learning for active vision

We develop a comprehensive description of the active inference framework, as proposed by Friston (2010), under a machine-learning compliant perspective. Stemming from a biological inspiration and the auto-encoding principles, a sketch of a cognitive architecture is proposed that should provide ways to implement estimation-oriented control policies under a POMDP perspective. Computer simulations illustrate the effectiveness of the approach through a foveated inspection of the input data. The pros and cons of the control policy are reviewed in details, showing interesting promises in term of processing compression, but also putative risks of a confirmation bias that may degrade the recognition performance if the model is too optimistic about its own predictions. The presented formalism is fully compliant with the auto-encoding framework and would deserve further developments under variational encoding architectures.