The methods developed in this thesis can be used for data analysis and are therefore useful as such. However, the research which has led to these methods has also been inspired by the biological brain. This section discusses some of the connections between the brain and the nonlinear factor analysis algorithm.
Although the biological brain does not implement any specific mathematically exact algorithm, Bayesian learning or others, the framework of Bayesian probability theory is appropriate for interpreting many of the different functions of the brain (see, e.g., [64]). Visual perception of shape from shading, for instance, was successfully analysed from the Bayesian point of view in [27].
On one hand, findings from the neurosciences can teach us how to build better algorithms for learning models from observations, and on the other hand, development of these algorithms can give valuable insight for interpreting the relevance of the findings. The model space and priors used by the brain, as well as the particular computational approximation can be interpreted to be implicitly defined by the genes because they give the brain the instructions on how to react to different environmental stimuli. The hypothesis space implicitly defined by the brain is evidently huge, computationally efficient and allows a rich representation of the environment.
It is obvious that learning is not the only function for the biological brain and explaining the structures found in the brain requires account to be taken of prediction and action. It seems plausible that the cerebral cortex is involved in both learning and inference or planning because the cortex is activated during imaging tasks [68]. However, the following concentrates only on learning. For textbook accounts on the brain see, e.g., [69,104,62].