TL
TL;DR
This work evaluates the progressive networks architecture extensively, and shows that it outperforms common baselines based on pretraining and finetuning and demonstrates that transfer occurs at both low-level sensory and high-level control layers of the learned policy.
Abstract
Learning to solve complex sequences of tasks--while both leveraging transfer and avoiding catastrophic forgetting--remains a key obstacle to achieving human-level intelligence. The progressive networks approach represents a step forward in this direction: they are immune to forgetting and can leverage prior knowledge via lateral connections to previously learned features. We evaluate this architecture extensively on a wide variety of reinforcement learning tasks (Atari and 3D maze games), and show that it outperforms common baselines based on pretraining and finetuning. Using a novel sensitivity measure, we demonstrate that transfer occurs at both low-level sensory and high-level control layers of the learned policy.
References24 items
1
A Deep Hierarchical Approach to Lifelong Learning in Minecraft
2
Asynchronous Methods for Deep Reinforcement Learning
3
Policy Distillation
4
Actor-Mimic: Deep Multitask and Transfer Reinforcement Learning
5