Inception Network Predicts Triple Intersection Numbers

This study uses an Inception convolutional neural network to predict triple intersection numbers of Calabi-Yau threefolds, reporting about 90% accuracy under standard fivefold cross-validation. The work moves beyond prior Hodge-number predictions by targeting intersection-theoretic invariants, a finer topological descriptor relevant to string compactifications. The results indicate that machine learning can learn refined algebraic-geometric invariants, potentially accelerating classification and computation in algebraic geometry.