IRM learns representations such that the optimal classifier is the same across training distributions, linking invariances to causal structures for improved out-of-distribution generalization.
Understanding deep learning requires rethinking generalization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2019 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Invariant Risk Minimization
IRM learns representations such that the optimal classifier is the same across training distributions, linking invariances to causal structures for improved out-of-distribution generalization.