Do Transformer World Models Give Better Policy Gradients?

A natural approach for reinforcement learning is to predict future rewards by unrolling a neural network world model, and to backpropagate through the resulting computational graph to learn a policy. However, this method often becomes impractical for long horizons since typical world models induce hard-to-optimize loss landscapes. Transformers are known to efficiently propagate gradients overlong horizons: could they be the solution to this problem? Surprisingly, we show that commonly-used transformer world models produce circuitous gradient paths, which can be detrimental …

computational cs.ai cs.lg future graph learn loss natural network neural network policy reinforcement reinforcement learning through transformer transformers world world models