This thesis aims at learning an Active Learning agent for one-shot predictions of texts and images via Reinforcement Learning. There are three different models being implemented, where the baseline is a LSTM network, and ...