Comparison of human and machine learning at image and text classification

OData support
Supervisor:
Dr. Szűcs Gábor
Department of Telecommunications and Media Informatics

Analytical thinking skill is determining part of our lives, we try to analyze our experiences and draw conclusions from it, thus we learn. Machines are capable of doing this learning process too. But how do machines perform compared to humans, if they learn under the same conditions?

In my thesis I have designed and implemented a test, which is suitable for the comparison of human and machine learning. The test is dealing with the problem of image and text classification. During the design process I have created manual test sets from collected image and text data. The test process was designed to analyze every single step of the learning, I made continuous (online) learning accordingly. This means, that after every classification the tester receives a feedback about the correctness of the latest decision. At start, both machine and human classifiers have equal chances, because there aren’t any further informations about the data, merely the visible content.

For the human testing it was necessary to have an adequate platform, where everyone can easily and swiftly fill out the tests. For that particular purpose, I have developed a website which was able to serve the test persons. I carried out simple machine classifiers for machine learning on text data, which can give the label of the unseen element based on the implemented algorithms.

I have evaluated the final results of the classifications for each test in details and I have compared the analyzed performance values. In this paper I have presented the necessary data for the best results.

Downloads

Please sign in to download the files of this thesis.