For learners restricted to ERM or weak consistency oracles, the paper proves exponential mistake and regret lower bounds in online learning and polynomial-query upper bounds in transductive online learning, with improved randomized algorithms for specific classes.
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
REJECT 1representative citing papers
citing papers explorer
-
Tradeoffs between Mistakes and ERM Oracle Calls in Online and Transductive Online Learning
For learners restricted to ERM or weak consistency oracles, the paper proves exponential mistake and regret lower bounds in online learning and polynomial-query upper bounds in transductive online learning, with improved randomized algorithms for specific classes.