Double preconditioning (DoPr) improves downstream task performance in test-time feedback settings without consistent gains in validation loss.
Sample efficient linear meta-learning by alternating minimization
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 2years
2026 2verdicts
UNVERDICTED 2roles
background 1polarities
background 1representative citing papers
CMTRL recovers a shared low-rank feature matrix for T constrained linear bandit tasks in d dimensions using Safe-AltGDmin and provides regret and sample complexity bounds.
citing papers explorer
-
Double Preconditioning (DoPr): Optimization for Test-Time Performance, not Validation Loss
Double preconditioning (DoPr) improves downstream task performance in test-time feedback settings without consistent gains in validation loss.
-
Multi-Task Representation Learning for Conservative Linear Bandits
CMTRL recovers a shared low-rank feature matrix for T constrained linear bandit tasks in d dimensions using Safe-AltGDmin and provides regret and sample complexity bounds.