The Effect of Noise on the Density Classification Task for Various Cellular Automata Rules

Area of Science:

• Artificial life
• Complex systems
• Computational science

Background:

• Cellular automata (CA) are discrete dynamical systems crucial for artificial life research.
• The density classification task (DCT) challenges CA to achieve global consensus from local information, mimicking quorum sensing.
• Previous studies assumed stable inputs, unlike real-world noisy biological systems.

Purpose of the Study:

• Investigate the impact of noise on the accuracy of high-performance CA rules for DCT.
• Analyze the trade-off between noise-free accuracy and robustness to noise.
• Compare performance of human-designed versus computationally evolved CA rules.

Main Methods:

• Simulated 1-dimensional Boolean CA on a ring lattice for the DCT.
• Introduced noise into the system to assess robustness.
• Utilized cubewalkers, a GPU-accelerated Boolean simulator, for large-scale experiments.

Main Results:

• A clear trade-off exists between maximum accuracy without noise and robustness to noise among top-performing CA rules.
• No significant performance difference was observed between human-designed and evolved CA rules.
• Noise significantly impacts the accuracy of CA rules previously considered highly accurate.

Conclusions:

• CA rule performance in DCT is sensitive to noise, revealing a critical trade-off.
• The origin of CA rules (designed vs. evolved) does not confer a significant advantage in noisy conditions.
• Future research should consider noise resilience when evaluating CA for complex tasks.