Roman Vlasov

Nikolay Sergievskiy, Roman Vlasov, Roman Trusov

Face anti-spoofing has drawn a lot of attention due to the high security requirements in biometric authentication systems. Bringing face biometric to commercial hardware became mostly dependent on developing reliable methods for detecting fake login sessions without specialized sensors. Current CNN-based method perform well on the domains they were trained for, but often show poor generalization on previously unseen datasets. In this paper we describe a method for utilizing unsupervised pretraining for improving performance across multiple datasets without any adaptation, introduce the Entry Antispoofing Dataset for supervised fine-tuning, and propose a multi-class auxiliary classification layer for augmenting the binary classification task of detecting spoofing attempts with explicit interpretable signals. We demonstrate the efficiency of our model by achieving state-of-the-art results on cross-dataset testing on MSU-MFSD, Replay-Attack, and OULU-NPU datasets.

Andrey Voronin, Roman Vlasov, Vladimir Gorgadze et al.

Constant Product Market Makers use fees that are typically fixed proportions of trade size. When these fees are automatically reinvested into the pool, as in Uniswap~V2 and some designs of Uniswap V4, the final state after a trade can depend on how the trade is split into smaller transactions. This path dependence complicates the risk assessment for liquidity providers and affects composability guarantees. We characterize the functional class of fee structures that ensure path independence: the combined fee factor must depend only on the current pool invariant k=xy. For this class, we derive a system of ordinary differential equations governing pool dynamics and obtain a closed-form integral exchange formula. Within this class, we construct a parametric family of fee functions that achieve zero Impermanent Loss for a given initial pool state, and prove that no universal fee function can eliminate Impermanent Loss for all initial states simultaneously. We analyze implications for arbitrage windows and slippage, and validate our theory through controlled simulations. Our framework provides protocol designers with a principled approach to fee optimization that aligns liquidity provider and trader incentives while preserving composability.

Ignat Melnikov, Roman Vlasov, Vladimir Gorgadze et al.

Decentralized Finance (DeFi) is a rapidly evolving segment of blockchain technology that enables a transformative approach to financial services through Web3 applications. By leveraging smart contracts, DeFi allows developers to build flexible and innovative financial instruments. Among the most prominent DeFi primitives by liquidity are decentralized exchange~(DEX) swap protocols~(such as Uniswap, Curve, and Balancer) that facilitate fast token-to-token exchanges. However, new exchange mechanisms also introduce new market inefficiencies that can be systematically exploited by arbitrageurs. This paper focuses on swap protocols based on the Automated Market Maker~(AMM), where the product of reserves is preserved as an invariant. We analyze the interaction between arbitrageurs and AMM liquidity pools and develop a mathematical model grounded in empirical pool configurations. Using this model, we derive bounds on the joint revenue of liquidity providers~(LPs) and arbitrageurs, propose a method to estimate the expected number of blocks until the occurrence of Impermanent Loss~(IL), and obtain a lower bound on the pool fee required to achieve a fixed target probability of staying in the Impermanent Gain (IG) zone within a block. The proposed framework extends existing LP risk-assessment methodologies by quantifying symbiotic profitability zones, providing a principled basis for fee selection that aligns LP-arbitrageur incentives and enhances market stability.

Dmitry Yashunin, Tamir Baydasov, Roman Vlasov

Currently in the domain of facial analysis single task approaches for face detection and landmark localization dominate. In this paper we draw attention to multi-task models solving both tasks simultaneously. We present a highly accurate model for face and landmark detection. The method, called MaskFace, extends previous face detection approaches by adding a keypoint prediction head. The new keypoint head adopts ideas of Mask R-CNN by extracting facial features with a RoIAlign layer. The keypoint head adds small computational overhead in the case of few faces in the image while improving the accuracy dramatically. We evaluate MaskFace's performance on a face detection task on the AFW, PASCAL face, FDDB, WIDER FACE datasets and a landmark localization task on the AFLW, 300W datasets. For both tasks MaskFace achieves state-of-the-art results outperforming many of single-task and multi-task models.