L-System vs. Matrix Neural Network Encoding: 24 Experimental Runs

Lsys genetic encoding achieved a 2.74x performance advantage over Matrix encoding in evolving neural networks within an artificial world. The dataset compares two genetic encoding methods across 24 runs, with results including food counts and generalization performance in a novel maze. Authored by Alexander Stuy and last updated on 2026-05-03.

Use Cases

• Benchmarking genetic encoding methods based on performance metrics like food count and coefficient of variation.
• Studying generalization of evolved neural networks based on transfer to a novel maze environment.
• Analyzing the impact of initial population structure on evolutionary performance based on the MatrixLSG control condition.

Strengths

• Includes quantitative results from 24 experimental runs, with specific performance metrics like a mean maximum food count of 3802 ± 197 for Lsys encoding.
• Demonstrates generalization capability, with Lsys populations achieving a 5.82x advantage in a novel maze environment.
• Provides a direct comparison between two genetic encoding methods (Lsys and Matrix) under controlled conditions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset is small (333.0 KB), indicating limited scope and likely containing summary results rather than raw evolutionary traces.

Provenance

Source

Alexander Stuy via figshare

Collection Method

Results from computational experiments comparing neural network genetic encoding methods.

Freshness

Last updated 2026-05-03 08:00:28; freshness should be verified.