A realizably learnable distribution class is proven not learnable against an adaptive additive adversary, while the same class is learnable against oblivious additive tampering.
On the power of adaptivity in statistical adversaries
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
-
On the Learnability of Distribution Classes with Adaptive Adversaries
A realizably learnable distribution class is proven not learnable against an adaptive additive adversary, while the same class is learnable against oblivious additive tampering.