If you had a person who could only learn by imitation, but could learn very sample efficiently, and could effectively recall and synthesize past knowledge, would you count that as continual learning?
For example, this person learns metalworking in a day by working under a blacksmith. Then they learn violinmaking from a luthier the next day. Then all on their own they can make a violin made of metal.
But they don’t do any online learning / exploration, just imitation.
I would claim this is still “true” continual learning in the ways it’s typically discussed. Do you agree?
If so, I think the bar is lower than your post makes it out to be: it would seem feasible that improvements in our current actual imitation learning algorithms could achieve continual learning.
If you had a person who could only learn by imitation, but could learn very sample efficiently, and could effectively recall and synthesize past knowledge, would you count that as continual learning?
For example, this person learns metalworking in a day by working under a blacksmith. Then they learn violinmaking from a luthier the next day. Then all on their own they can make a violin made of metal.
But they don’t do any online learning / exploration, just imitation.
I would claim this is still “true” continual learning in the ways it’s typically discussed. Do you agree?
If so, I think the bar is lower than your post makes it out to be: it would seem feasible that improvements in our current actual imitation learning algorithms could achieve continual learning.