Towards Learning Knot Invariants using Graph Neural Networks
Published in n/a, 2023
So far, very little research has been done on the applications of machine learning to the topological field of knot theory. To the best of our knowledge, we are the first to investigate directly leveraging the topological structure of knots for learning tasks to predict certain properties. We show that graph neural networks, in particular graph isomorphism networks, provide better accuracy in predicting the symmetry of prime knots, indicating that the structure of knots can indeed be leveraged. While the performance of models is still not sufficient for useful applications in research in mathematics, we hope that this study inspires further research at the intersection of knot theory and geometric learning methods.