1: \begin{abstract}

2: 

3: This work studies minimization problems with zero order noisy oracle information under assumption that the objective function is highly smooth and, possibly satisfies additional conditions. 

4: We present an improved analysis of the algorithm of~\cite{BP2016} \massi{out of the blue. We should first introduce first the general algorithm, which is considered in many papers, with the precise version considered in this paper due to BP2016}in the case of strongly convex functions for constrained and unconstrained problems.

5: For the same algorithm, we derive rates of convergence without additional assumptions and with the Polyak-Łojasiewicz condition instead of the strong convexity.

6: We propose a new algorithm, which is based on $\ell_1$-type randomization in place of the $\ell_2$-one utilized by~\cite{BP2016}.

7: This new algorithm enjoys similar theoretical guarantees and, in case of noiseless oracle, enjoys better bounds on its bias and variance.

8: For both algorithms, we analyse the case of quadratic, but not necessary strongly convex, functions, establishing improved rates of convergence.

9: The improvements are achieved by new bounds on bias and variance for both algorithms, which are obtained via Poincaré type inequalities for uniform distributions on $\ell_p$ spheres with $p \in \{1, 2\}$.

10: The optimality of the upper bounds is discussed and a slightly more general lower bound is proposed. 

11: \end{abstract}

12: