Symbolism and Visual Information Ontology from Connectionist Artificial Intelligence

The paper presents a framework for structural semantic ordering predicated on visual process understanding from intelligent machines. This framework operates on high-level features and class membership probabilities that are extracted from canonical stochastic fields by these machines. The proposed contributions fall into the following categories. First, the paper reveals insights into how the equivalent convolution tensor filter sharpens the information penalization process in deep layers of the machines under consideration. Second, it highlights how the dense neuron matrix parameterization can exhibit correlation patterns associated with class ordering from the template training dataset of these machines. Third, it emphasizes how determining probabilistic categorical centroids based on softlog softmax divergence can reveal surprising behaviors of advanced computer vision machines. Lastly, the paper proposes a structural semantic class/entity/object ordering derived from the responses of machines to requests received in order to force the acceptance of an equivalence between structured entities and simple/unstructured stochastic processes. This ordering process, derived from vision machines in the sense of divergence from simplicity, is called visual dissimplicity. Is a tiger visually simpler to describe than an oscilloscope? Some connectionist machines will seek to answer this question.