Generative Hybrid Representations for Activity Forecasting With No-Regret Learning | IEEE Conference Publication | IEEE Xplore