Author name cluster

Yuriy Dorn

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

3 papers

1 author row

AAMAS Conference 2026 Conference Paper

Functional Multi-armed Bandit and the Best Function Identification Problems

Yuriy Dorn
Aleksandr Katrutsa
Ilgam Latypov
Anastasia Soboleva

We consider the model selection problem, where we have a set of candidate parametric functions and need to identify the function with the smallest minimum and corresponding minimizer. This problem arises in the competitive training of neural networks, where a set of candidates is given, and the limited computational budgetpreventstheuseofabrute-forcesearch. Toaddressthisproblem, we propose generalizations of the classical multi-armed bandit (MAB)andbestarmidentification(BAI)setups, sinceusingclassical MAB and BAI setups leads to infeasible computational costs. We refer to the proposed setups as the functional multi-armed bandit problem (FMAB) and the best function identification (BFI) problems, respectively. For these problems, we establish lower regret bounds for different classes of candidate functions. To solve FMAB and BFI problems, we propose a novel reduction scheme to construct the F-LCB algorithm, which is a UCB-type algorithm based on basic algorithms for nonlinear optimization with known convergence rates. TheF-LCBalgorithmcombinesthearmselectionstepandthe update of the current optimum approximation. We provide regret upper bounds for F-LCB based on the known convergence rates of the underlying base algorithms. The regret upper bounds match with the derived lower bounds up to the logarithmic factor. Numerical experiments confirm that the proposed approach correctly identifies the optimal function and provides the minimizer for it in both smooth and non-smooth convex cases. Similarly, F-LCB converges faster than SuccessiveHalving and Hyperband algorithms for the model selection problem, where the candidate functions are neural networks and only a stochastic gradient estimate is available.

PDF

AAMAS Conference 2026 Conference Paper

Robust Autobidding for Noisy Conversion Prediction Models

Andrey Pudovikov
Alexandra Khirianova
Ekaterina Solodneva
Gleb Molodtsov
Aleksandr Katrutsa
Yuriy Dorn
Egor Samosvat

Managing millions of digital auctions is essential to modern advertising auction systems. The primary approach to managing digital auctions is autobidding, which relies on Click-Through Rate and Conversion Rate metrics. While these quantities are estimated with ML models, their prediction uncertainty directly impacts advertisers’ revenue and bidding strategies. To address this issue, we propose RobustBid, an efficient method for robust autobidding taking intoaccountuncertaintyinCTRandCVRpredictions. Ourapproach leveragesadvanced, robustoptimizationtechniquestopreventlarge errors in bids if the estimates of CTR/CVR are perturbed. We derive an analytical solution to the stated robust optimization problem, which improves the runtime efficiency of the RobustBid method. The synthetic, iPinYou, and BAT benchmarks are used in our experimental evaluation of RobustBid. We compare our method with the non-robust baseline and the RiskBid algorithm using total conversion volume (TCV) and average cost-per-click (𝐶𝑃𝐶𝑎𝑣𝑔) as performance metrics. The experiments demonstrate that RobustBid provides bids that yield larger TCV and smaller𝐶𝑃𝐶𝑎𝑣𝑔 than competitors in the case of large perturbations in CTR/CVR predictions.

PDF

AAMAS Conference 2025 Conference Paper

Fast UCB-type Algorithms for Stochastic Bandits with Heavy and Super Heavy Symmetric Noise

Yuriy Dorn
Aleksandr Katrutsa
Ilgam Latypov
Andrey Pudovikov

This paper considers stochastic multi-armed bandit problems (MAB) and presents a novel framework for constructing UCB-type algorithms. The main ingredient of UCB-type algorithms is the estimate of the confidence bound typically derived from statistical assumptions. On the opposite, our approach derives the confidence bounds from the convergence rate of the base convex optimization method, which helps to solve auxiliary optimization problems in every round. To show the relations between the convergence of the optimization method and the novel UCB-type algorithm, we derive the regret bounds corresponding to the convergence rates of the selected optimization method. To illustrate the proposed framework, we introduce a new algorithm, Clipped-SGD-UCB, for the MAB with heavy-tailed reward distribution, where Clipped-SGD is used as a base convex optimization method since its convergence for the heavy-tail inexact oracle is known. We show theoretically and empirically that in the case of symmetric noise in the reward distribution, one can achieve an𝑂(log𝑇 √︁ 𝐾𝑇 log𝑇) regret bound instead of 𝑂 𝑇 1 1+𝛼 𝐾 𝛼 1+𝛼. These bounds correspond to the cases where the reward distribution satisfies E𝑋 ∈D [|𝑋 |1+𝛼 ] ≤ 𝜎1+𝛼 (𝛼 ∈ (0, 1]), i. e. perform better than it is assumed by the general lower bound for bandits with heavy-tails.

PDF