Adapting a “maple-leaf vs oak-leaf” net to “beech vs birch” is like…

Question

Adapting a “maple-leaf vs oak-leaf” net to “beech vs birch” is likely to work. Asking that same net to score recorded bird-call accents is not. What should the tester summarize?

Accepted Answer

Fine-tuning success tracks similarity between the original task and the new one. Small task distance supports fine-tuning; large distance does not. — Adapting maple-versus-oak to beech-versus-birch is likely to work. Asking that same net to score bird-call accents is not. Fine-tuning success tracks similarity between the original task and the new one. Small task distance supports it; large distance does not.

Answer

Fine-tuning always works, because any pretrained net transfers to any new task.

Answer

Fine-tuning never works when the new task is a nearby leaf pair.

Answer

A Chapter 5 representativeness procedure as the only transfer rule.

Adapting a “maple-leaf vs oak-leaf” net to “beech vs birch” is likely to work. Asking that same net to score recorded bird-call accents is not. What should the tester summarize?

Short Explanation

Full Explanation