We present our development process of an algorithm to confidently classify costly sequential input while utilizing minimal but sufficient data acquisitions. More specifically, we solve the binary classification of articles by subject, using minimal number of sentences, determining stop condition using Sequential Probability Ratio Test (SPRT) as a measure.
This method can be extremely useful in a variety of tasks, particularly in the context of the state-of-the-art deep learning branch of the machine learning field – where variations of the classification problems are one of the hottest challenges at stake. In this project we implemented the SPRT on a CNN architecture for sentence classification.
Dr. Alon Amar RAFAEL