GEAKG: Generative Executable Algorithm Knowledge Graphs

Dimensions overall score 7.0

• GEAKG is a domain-agnostic framework where the same three-layer architecture and Ant Colony Optimization-based learning engine can be instantiated across different problem domains without framework-level code changes.method 90%
  Evidencepartial

  The framework is domain-agnostic at the engine level: the same three-layer architecture and Ant Colony Optimization (ACO)-based learning engine can be instantiated across domains, parameterized by a pluggable ontology (RoleSchema).

  Implicationpartial

  Directly and explicitly stated in the abstract and analysis excerpt as a core framework hypothesis.

  Verificationpartial

  partial

• GEAKG enables zero-shot transfer of learned algorithmic composition patterns across domains, meaning deployment on a target domain uses a fixed snapshot without additional retraining.result 90%
  Evidencepartial

  Transferable: Meta-level knowledge persists and transfers zero-shot across domains

  Implicationpartial

  Explicitly defined in the analysis excerpt and presented as a key property in the abstract and results.

  Verificationpartial

  partial

• GEAKG nodes store executable operators (runnable code), making the knowledge graph directly executable.technical 90%
  Evidencepartial

  A GEAKG is generative (topology and operators are synthesized by a Large Language Model), executable (every node is runnable code), and transferable (learned patterns generalize zero-shot across domains).

  Implicationpartial

  Explicitly stated as a defining characteristic of GEAKG in the abstract.

  Verificationpartial

  partial

• The topology and operators of a GEAKG are synthesized by a Large Language Model, making the graph generative.method 90%
  Evidencepartial

  A GEAKG is generative (topology and operators are synthesized by a Large Language Model)...

  Implicationpartial

  Directly stated as a defining characteristic in the abstract and elaborated in the analysis.

  Verificationpartial

  partial

• GEAKG's learned knowledge (L2) enables transfer without requiring LLM calls at runtime, reducing inference costs compared to pure LLM-based code evolution approaches.technical 85%
  Evidencepartial

  The learned knowledge (L2) enables transfer without requiring LLM calls at runtime.

  Implicationpartial

  Strongly implied in the analysis excerpt as a key advantage over methods like LLaMEA, which incur high inference costs.

  Verificationpartial

  partial

• GEAKG is distinct from AutoML systems because it represents procedural knowledge as an executable graph, rather than optimizing pipeline configurations.technical 90%
  Evidencepartial

  A GEAKG is not an AutoML system. AutoML optimizes pipeline configurations (model selection, hyperparameters); a GEAKG represents procedural knowledge as an executable graph.

  Implicationpartial

  Explicitly stated in a dedicated comparison section within the provided text.

  Verificationpartial

  partial

• GEAKG enforces schema-constrained composition and executes only offline-validated operators, allowing the LLM to act as an architect rather than a coder.method 85%
  Evidencepartial

  This enforces schema-constrained composition and executes only offline-validated operators at runtime, letting the LLM act as architect rather than coder.

  Implicationpartial

  Directly stated in the analysis excerpt as a core design mechanism.

  Verificationpartial

  partial

• Current knowledge graph paradigms provide limited support for representing procedural knowledge as executable, learnable graph structures.limitation 80%
  Evidencepartial

  yet current KG paradigms provide limited support for representing procedural knowledge as executable, learnable graph structures.

  Implicationpartial

  Directly stated as a motivation for the work in the abstract, though it is a general claim about the field.

  Verificationpartial

  partial

Competitive landscape

A framework for learning and transferring algorithmic expertise as executable knowledge graphs, enabling zero-shot generalization across diverse problem domains.