|Title||Evaluation of sum-product networks on image classification tasks|
|Publication Type||Journal Article|
|Year of Publication||2015|
|Authors||Bonanno, D, Bhattacharya, S, Smith, L, Aha, DW|
We present a method for classifying imagery using Sum-Product Networks. Current techniques allow for the architecture to be learned in addition to learning the weights between nodes, resulting in high accuracy classification without the need for manual architecture specifications. Our results show that such networks can perform comparably to current state of the art methods on simple image classification problems. However, applying SPNs to this task requires substantially reducing the sizes of the images, and learning the structure of an SPN for the originally-sized images would be computationally expensive.
NRL Publication Release Number: