ΔPPL Is Not ΔCE: The Scale-Dependence Confound in Generative Continual-Learning Forgetting Metrics
This paper demonstrates that reporting forgetting in generative continual learning as raw perplexity change (ΔPPL) introduces a scale-dependent confound that artificially inflates or deflates differential forgetting ratios based on base perplexity levels, arguing instead for cross-entropy (ΔCE) as a scale-invariant metric to reveal the true underlying model drift.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.