Gives a poly-time testable learner for halfspaces with Gaussian marginals and Massart noise, plus a super-polynomial lower bound separating testable learning from classical learning under random classification noise.
From the run-time guarantees given in Theorem B.1 and Theorem C.1, we see immideately that the run-time of the algorithmA Massart ispoly d ϵ log 1 δ
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Testing Noise Assumptions of Learning Algorithms
Gives a poly-time testable learner for halfspaces with Gaussian marginals and Massart noise, plus a super-polynomial lower bound separating testable learning from classical learning under random classification noise.