This thesis introduces a new approach to Glacial Isostatic Adjustment (GIA) modeling using Machine Learning (ML) techniques. The work addresses two main challenges – uncertainty in historical ice load history and the complexity of inverse problems – by developing two ML-based surrogate models (emulators) to rapidly estimate Relative Sea-Level (RSL) history and uplift rates from varying ice load histories, given a constant Earth model. The emulators are constructed using a Convolutional Neural Network (CNN) with U-Net architecture and spherical data representation, enabling an efficient, cost-effective approach to the GIA forward model. The performance of these emulators was evaluated in two separate experiments, displaying encouraging results in efficiency, accuracy, and versatility. The prediction performance exceeded existing models in computational speed and offers accuracy comparable to other GIA studies. The successful implementation of these emulators could advance GIA modeling by integrating ML, but enhanced resolution is needed for direct scientific application.