1: \begin{abstract}

2: We address the problem of \emph{convex optimization with dueling feedback}, where the

3: goal is to minimize a convex function given a weaker form of \emph{dueling} feedback. Each query consists of two points and the dueling feedback returns a (noisy) single-bit binary comparison of the function values of the two queried points.

4: The translation of the function values to the single comparison bit is through a \emph{transfer function}.

5: % 

6: %This is unlike the traditional optimization setting which assumes an oracle access to gradient or function value information. 

7: %

8: This problem has been addressed previously for some restricted classes of transfer functions, but here we consider a very general transfer function class which includes all functions that 

9: % can be approximated by a finite polynomial with a minimal degree $p$.

10: admit a series expansion with minimal degree $p$ about the origin.

11: %degree-$p$ polynomial link function based preference class as the underlying dueling-feedback model. %

12: %

13: Our main contribution is an efficient algorithm with $\epsilon$-optimal convergence rate of $\smash{O}(\epsilon^{-4p})$ for smooth convex functions, and an optimal rate of $\smash{\widetilde O}(\epsilon^{-2p})$\footnote{$\tilde O(\cdot)$ notation hides logarithmic factors.} when the objective is both smooth and strongly convex.

14: % 

15: % Our work closes the open problem of function optimization with general polynomial transfer functions, beyond the sign feedback and strong convexity assumptions.

16: %

17: %The efficacy of our proposed algorithms are justified through empirical comparisons with an existing baseline. % which shows them to outperform the state of the art.

18: \end{abstract}

19: