New AI Learns Without Forgetting Previous Skills
This article discusses a new AI technique called Progressive Neural Networks that can learn new tasks without forgetting previous skills. The key is adding new parts that connect to old ones, allowing the system to reuse prior knowledge.
Why it matters
This AI technique could enable the development of smart helpers that get better at assisting humans as they gain more experience.
Key Points
- 1Progressive Neural Networks avoid forgetting old skills when learning new tasks
- 2The system adds new parts that connect to previous ones, allowing reuse of prior knowledge
- 3It performs better than traditional train-and-tweak methods on simple video games and 3D mazes
- 4The approach enables machines to learn over time without losing past lessons
Details
The Progressive Neural Networks approach aims to enable AI systems to continuously learn new tasks and skills without forgetting what they previously knew. Instead of erasing old knowledge when learning new things, the system adds new parts that connect to the existing neural network. This allows the AI to reuse prior knowledge and build on top of its previous capabilities when facing new problems. Researchers have tested this on simple video games and 3D maze tasks, finding that it tends to learn faster and retain earlier skills better than traditional train-and-tweak methods. The key is the side connections that share useful features, so new skills build on old ones rather than destroying them. This points the way towards AI assistants that can improve with experience over time, without losing past lessons.
No comments yet
Be the first to comment