Abrar Ahmed, Anish Bikmal

We propose a computational model that can successfully learn an objectcategory from as few as one example and allows its learning style to betailored explicitly to a scenario. Our model decomposes each image into twoattributes: shape and color distribution. We then use a Bayesian criterion toprobabilistically determine the likelihood of each category. The model takeseach factor into account based on importance and calculates the conditionalprobability of the object belonging to each learned category. Our model is notonly applicable to visual scenarios, but can also be implemented in a broaderand more practical scope of situations such as Natural Language Processing aswell as other places where it is possible to retrieve and construct individualattributes. Because the only condition our model presents is the ability toretrieve and construct individual attributes such as shape and color, it can beapplied to essentially any class of visual objects.

location

arxiv.org

Advertisements