Predicting Cellular Branching with Inverse Cellular Automata via Recurrent Neural Network

Abstract

Biological processes are very complex so the predicting of such processes is considered to be very difficult. When it comes to the image-based prediction, one of the main reason of the predicting difficulty comes from the cost that gathering enough visual data which is extremely expensive. In this thesis, we introduce a novel method to reduce the cost of the prediction of cellular branching with Inverse Cellular Automata (ICA). This method assumes the biological image sequence as the set of cellular automata rules between t and t+1. With this assumption, single biological image sequence can provide width * height * frame number of cellular automata set of rule samples so they can be used as traditional Recurrent Neural Network(RNN) based machine learning. Proposed method overcomes the problem of lack of samples for machine learning in biological image predicting. We first briefly review the traditional cellular automata and introduce the concept of the ICA. And we will check the learnability with existing automata image rules. Finally, we show the prediction images with proposed method.