Core Concepts
The authors introduce the Phase Transition Finder (PTF) algorithm to efficiently generate parameters at phase boundaries, aiming to discover complex behaviors in continuous systems like Lenia.
Abstract
In the quest for interesting life-like behaviors in Artificial Life systems, the authors propose an automated approach with PTF. By focusing on phase transition regions, they aim to increase the frequency of intriguing dynamics while maintaining scalability and efficiency. The method involves defining phases, sampling transition points, and exploring parameter space to uncover complex behaviors.
Stats
In continuous systems like Lenia, only a small subset of parameter space displays interesting dynamics.
Estimations suggest that outer-holistic 2D cellular automata have approximately 1% interesting dynamics.
Multi-channel Lenia has 106 free parameters influencing its dynamics.
The PTF algorithm aims to efficiently explore phase transition regions between different phases.
Results show that points in phase transition regions exhibit a higher percentage of interesting configurations compared to random sampling.
Quotes
"In this paper, we explore a complementary method to generate interesting systems at a much higher rate than random sampling." - Papadopoulos, Doat, Renard, Hongler
"The main idea we seek to exploit is the observation that phase transition regions often display complex dynamics." - Papadopoulos, Doat, Renard, Hongler
"Overall, qualitatively we observe that for phase transition points, the presence of solitons, static or even moving is not uncommon." - Papadopoulos, Doat, Renard, Hongler