Different from direct reconstruction from fMRI signals we transferred the understanding of brain activity into the understanding of feature representations in CNN by by training a mapping from fMRi signals to hierarchical features extracted from CNN

Different from direct reconstruction from fMRI signals we transferred the understanding of brain activity into the understanding of feature representations in CNN by by training a mapping from fMRi signals to hierarchical features extracted from CNN.
The deep generative model implemented the perceived image reconstruction problem and its helps to derive the predictive distribution to reconstruct the visual images from brain activity and also deal with encoding tasks. This method has high computational complexity.

Whereas, Linear reconstruction model, provide the high quality of reconstruction from the stimuli, by inverting the properly encoding reconstruction model. During encoding and decoding task, the performances were analyzed with regression.

In order to perform the constraint-free visual image reconstruction, discriminant functions are used, which deals with the reconstruction of the visual image at multiple scales. This method provides information about the activity which is discovered from the human brain activity.
The recent method, DNN feature decoding was used to minimize the cost function of layers using the algorithm of LMBFGS, and it has solved the non-optimization problem on reconstruction with unconstraint values.