Pith. sign in

High-Probability Risk Bounds via Sequential Predictors

2 Pith papers cite this work. Polarity classification is still indexing.

2 Pith papers citing it
abstract

Online learning methods yield sequential regret bounds under minimal assumptions and provide in-expectation risk bounds for statistical learning. However, despite the apparent advantage of online guarantees over their statistical counterparts, recent findings indicate that in many important cases, regret bounds may not guarantee tight high-probability risk bounds in the statistical setting. In this work we show that online to batch conversions applied to general online learning algorithms can bypass this limitation. Via a general second-order correction to the loss function defining the regret, we obtain nearly optimal high-probability risk bounds for several classical statistical estimation problems, such as discrete distribution estimation, linear regression, logistic regression, and conditional density estimation. Our analysis relies on the fact that many online learning algorithms are improper, as they are not restricted to use predictors from a given reference class. The improper nature of our estimators enables significant improvements in the dependencies on various problem parameters. Finally, we discuss some computational advantages of our sequential algorithms over their existing batch counterparts.

years

2026 2

representative citing papers

Efficient Logistic Regression with Mixture of Sigmoids

cs.LG · 2026-04-03 · unverdicted · novelty 7.0

EW with Gaussian prior matches the optimal O(d log(Bn)) regret for online logistic regression at O(B^3 n^5) cost and converges geometrically to a truncated Gaussian vote in the large-B separable regime.

citing papers explorer

Showing 2 of 2 citing papers.

  • Gradient-free stochastic optimization of derivatives under strong convexity math.ST · 2026-07-08 · accept · none · ref 55 · internal anchor

    The minimax optimal rate for minimizing the k-th derivative of a Hölder function from noisy zero-order queries is N^{-(β-1)/(β+k)}, achieved by a kernel-based projected stochastic gradient algorithm.

  • Efficient Logistic Regression with Mixture of Sigmoids cs.LG · 2026-04-03 · unverdicted · none · ref 41

    EW with Gaussian prior matches the optimal O(d log(Bn)) regret for online logistic regression at O(B^3 n^5) cost and converges geometrically to a truncated Gaussian vote in the large-B separable regime.