Scaling All-Goals Updates in Reinforcement Learning Using Convolutional Neural Networks