Repeated maximum-likelihood retraining on self-generated data collapses exponential-family models, but a single fresh external data point or a prior stabilizes them.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Lost in Retraining: Roaming the Parameter Space of Exponential Families Under Closed-Loop Learning
Repeated maximum-likelihood retraining on self-generated data collapses exponential-family models, but a single fresh external data point or a prior stabilizes them.