Forecasting Surface Velocity Fields Associated With Laboratory Seismic Cycles Using Deep Learning