Visual Graph Arena Dataset

Tasks

The Visual Graph Arena Benchmark is a collection of six datasets designed to evaluate and enhance the visual reasoning capabilities of AI. The datasets are structured around three primary concepts, each divided into two tasks:

1. Graph Isomorphism Tasks

These tasks challenge AI systems to identify structural equivalence between graphs, regardless of their visual representation. (Labels: True/False)

Easy Isomorphism

Task: Determine whether two given graphs are isomorphic, with non-isomorphic graphs chosen randomly. This tests basic graph comparison skills.

Hard Isomorphism

Task: Determine whether two given graphs are isomorphic, with non-isomorphic graphs being degree-equivalent. This advanced version tests deeper structural analysis capabilities. (Labels: True/False)

2. Graph Path Finding Tasks

These tasks evaluate an AI's ability to navigate graph structures, find optimal routes, and understand connectivity within networks.

Hamiltonian Path

Task: Determine whether a given graph contains a Hamiltonian path (a path that visits each vertex exactly once). This tests the AI's ability to explore all possible paths efficiently.

Shortest Path

Task: Find the shortest path length between two marked nodes in a graph. This evaluates the AI's ability to optimize routes in complex networks.

3. Graph Cycle Finding Tasks

These tasks focus on identifying specific cyclic structures within graphs. The ability to detect and analyze cycles in a graph.

Hamiltonian Cycle

Task: Determine whether a given graph contains a Hamiltonian cycle (a cycle that visits each vertex exactly once and returns to the start). This tests the AI's ability to find complex cyclic patterns.

Biggest Chordless Cycle

Task: Find the size of the largest chordless cycle (a cycle with no shortcuts) in a graph. This evaluates the AI's ability to identify specific substructures within complex graphs.