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.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Continual Learning | Sketch (Fine-grained 6 Tasks) | Accuracy | 76.35 | ProgressiveNet |
| Continual Learning | Stanford Cars (Fine-grained 6 Tasks) | Accuracy | 89.21 | ProgressiveNet |
| Continual Learning | CUBS (Fine-grained 6 Tasks) | Accuracy | 78.94 | ProgressiveNet |
| Continual Learning | Wikiart (Fine-grained 6 Tasks) | Accuracy | 74.94 | ProgressiveNet |
| Continual Learning | ImageNet (Fine-grained 6 Tasks) | Accuracy | 76.16 | ProgressiveNet |
| Continual Learning | Flowers (Fine-grained 6 Tasks) | Accuracy | 93.41 | ProgressiveNet |