In a two-task linear regression model with random orthogonal tasks, sufficiently high overparameterization keeps the first task's risk low after training on the second task, and the forgetting ratio vanishes as the number of features grows.
Understanding double descent requires a fine-grained bias-variance decomposition
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Analysis of Overparameterization in Continual Learning under a Linear Model
In a two-task linear regression model with random orthogonal tasks, sufficiently high overparameterization keeps the first task's risk low after training on the second task, and the forgetting ratio vanishes as the number of features grows.