Scalability Property in Solving the Density Classification Task

Abstract

The density classification task (DCT for short) is one of the most studied benchmark problems for analyzing emergent computations performed by cellular automata. Starting from the observation that the performance of the best known solutions is not stable towards initial configurations size; we propose in this paper, some new evolutionary mechanisms for designing new solutions with similar conceptual properties to the best known ones. The approach is based on varying the size of initial configurations which allows making comparisons and analysis between the different solutions. We show then through a set of numerical results that the proposed mechanism allows collecting solutions for the DCT more efficiently and with reduced efforts. Also, we show that the collected solutions are affected by configurations size variations, where only few of them are scalable.

Combining Domain Knowledge Extraction With Graph Long Short-Term Memory for Learning Classification of Chinese Legal Documents

“…It is a classification task to predict the sentence of legal trial documents and recommend similar cases to judicial officials [42]. Graph LSTM is employed to classify legal documents based on sentencing in the paper.…”