Abstract:This work computationally investigates an algorithm that was originally proposed for creating and searching trees of patterns under the assumption that it seeks the most efficient internal representation of the external world, namely a real-world dataset of patterns. The algorithm produces a transition from star graph to regular 1d lattice, passing through irregular scale-invariant trees, under the effect of a single control parameter. The relatively novel information-theoretic tools of the von Neumann entropy and its derivative, the specific heat, together with the recent definition of thermodynamic quantities, such as work and heat, for networks are reviewed and applied to inspect the tree structures produced by the formation protocol. We observe some consistency between the optimal thermodynamic efficiency for networks and the optimal pattern-matching efficiency, defined in terms of the trade-off between accuracy and computational cost when the former is measured at the meso-resolution scale involved in the phase transition from star graph to scale-invariant network. We adduce the hypothesis of a richer frequency content at the meso-scale of the self-organized network to explain this observation. Moreover, empirical evidence suggests that the system benefits from the vicinity to the mentioned phase transition if it is faced with higher-complexity environments, while it does not when the environment is simpler. A phase portrait of the tree-generating algorithm is provided and multiple phase transitions are identified.

A Comparison of Empirical Tree Entropies

Entropy in Different Text Types.

Entropy, Derivation Operators and Huffman Trees

On Random Tree Structures, Their Entropy, and Compression

A Comparative Analysis of Discrete Entropy Estimators for Large-Alphabet Problems

Comparing entropies within the chinese language

Constructing Phylogenic Trees Based on Relative Entropy Theory

Entropic aspects of random fields on trees

An Empirical Comparison of Probability Models for Dependency Grammar

Tree shift topological entropy

Comparison study of using semantic and syntactic network characteristics to do text clustering

Comparison results for Markov tree distributions

Topological Entropy and Sequence Entropy for Hom Tree-Shifts on Unexpandable Trees

Unifying Attribute Splitting Criteria of Decision Trees by Tsallis Entropy.

Efficient Similarity Search for Tree-Structured Data

Tree networks of real-world data: analysis of efficiency and spatiotemporal scales

Discriminative Measures for Comparison of Phylogenetic Trees

Topological Entropy of Transitive Maps of a Tree

Entropy for $k$-trees defined by $k$ transition matrices

Entropy on regular trees

New ordinal relative fuzzy entropy