Train Short, Act Long: Multistability and the Horizon Independence of Recurrent Policies
The research investigates temporal horizon generalization (THG) in agents using recurrent neural networks for partially observable Markov decision processes (POMDPs). It finds that while parallel architectures are monostable and fail to generalize, nonlinear RNNs exhibiting multistability can achieve THG. The study concludes that both transient dynamics and multistability are essential for solving complex, long-horizon tasks.