Silver, Yang and Li define lifelong machine learning in [1] as: "Lifelong Machine Learning, or LML, considers systems that can learn many tasks over a lifetime from one or more domains. They efficiently and effectively retain the knowledge they have learned and use that knowledge to more efficiently and effectively learn new tasks."
Taylor, Kuhlmann and Stone name the requirements for a transfer learning algorithm in [2]: "1. Select an appropriate source task from which to transfer, given a target task. 2. Learn how the source task and target task are related. 3. Effectively transfer knowledge from the source task to the target task."
Aren't those pretty similar definitions?
[1] https://www.researchgate.net/publication/242025320_Lifelong_Machine_Learning_Systems_Beyond_Learning_Algorithms
[2] https://www.researchgate.net/publication/200502413_Autonomous_transfer_for_reinforcement_learning
Conference Paper Lifelong Machine Learning Systems: Beyond Learning Algorithms
Conference Paper Autonomous transfer for reinforcement learning
I will tend to say no, but just basing myself on your excerpts.
In the second definition I do not see the lifelong problem at the stake, nor the idea that you can improve your abilities to do something algorithmically, in automatic ways (machine learning). Also, the second definition is almost an algorithm, a method specified in some way about transferring knowledge. For example the ability to solve the Pythagorean theorem. But not a method, for example, to learn how to solve other theorems once you have learned to solve the Pythagorean theorem.
The first is more philosophical as a definition, or perhaps in some way a kind of open question (can we?), and is more focused on the issue of improving knowledge, long life, automatically, through the ML.
They appear to be a little different in these parts, to me.
- Badges
- Science topic