Abstract: To enhance the prediction accuracy and design efficiency for the performance of triangular corrugated sandwich beams under low-velocity impact, this study proposes a data-driven prediction method based on Multi-Task Learning (MTL). A sample dataset was constructed using a finite element model validated by experiments, and this dataset was used to train an MTL model for simultaneously predicting the structure's Specific Energy Absorption (SEA), maximum top plate deflection (<italic>δ</italic>_max), and initial peak load (<italic>F</italic>_max). The results show that the model achieves high prediction accuracy, with the coefficient of determination (R²) for each output response being no less than 0.99, effectively serving as a substitute for time-consuming finite element simulations. Further parameter sensitivity analysis reveals that core unit cell count (<italic>n</italic>) and web thickness of the core (<italic>t</italic>_c) have the most significant influence on structural stiffness, followed by the top plate thickness (<italic>t</italic>₁), while the bottom plate thickness (<italic>t</italic>₂) has the least impact. Additionally, a saturation threshold for performance improvement was observed for the web thickness of the core (<italic>t</italic>_c). Using the Non-dominated Sorting Genetic Algorithm II (NSGA-II), multi-objective optimization was conducted separately for deformation behavior, energy absorption performance, and comprehensive performance, resulting in optimal configurations that meet different design priorities.