Abstract: Determining the dynamics of the expectation values of operators acting on quantum many-body systems is a challenging task. Matrix product states (MPS) have traditionally been the ”go-to” models for these systems because calculating expectation values in this representation can be done with relative simplicity and high accuracy. However, such calculations can become computationally costly when extended to long times. Here, we present a solution for extending the computation of expectation values to long time intervals. We utilize a convolutional neural network model as a tool for the extended prediction of MPS generated expectation values calculated within the regime of short time intervals. With this model, the computational cost of generating long-time dynamics is significantly reduced, while maintaining reasonable accuracy. These results are demonstrated with operators relevant to quantum spin models in one spatial dimension.