Topic

Automatic Image Classification Supported by Expert Knowledge

Sergio Peñafiel

 

S.Peñafiel, B. Panay, N. Baloian, J. A. Pino, and W. Luther

Universidad de Chile

 

Abstract:

 

 In recent years, several algorithms have been proposed to solve image classification problems mainly based on convolutional neural networks which have reached high levels of accuracy. However, new studies state that these models are unstable, very sensitive to noise and requires many data samples to train. We propose a two-phase method to perform expert assisted image classification, in the first steps experts define constraints for the problem to look only at the important aspects of the image, in the second phase we trained a model using the values of the previous phase as input to perform interpretable classification.

 

 

 

Discussion Room: Automatic image classification supported by expert knowledge

 

[email protected]