arXiv subjects

Dmitry Korolev

Publications and source records attributed to Dmitry Korolev.

4 recordsLinked to original sources

arXiv2025

LATTE: Learning Aligned Transactions and Textual Embeddings for Bank Clients

Learning clients embeddings from sequences of their historic communications is central to financial applications. While large language models (LLMs) offer general world knowledge, their direct use on long event sequences is computationally expensive and impractical in real-world pipelines. In this paper, we propose LATTE, a contrastive learning framework that aligns raw event embeddings with semantic embeddings from frozen LLMs. Behavioral features are summarized into short prompts, embedded by the LLM, and used as supervision via contrastive loss. The proposed approach significantly reduces inference cost and input size compared to conventional processing of complete sequence by LLM. We experimentally show that our method outperforms state-of-the-art techniques for learning event sequence representations on real-world financial datasets while remaining deployable in latency-sensitive environments.

cs.CL↗

arXiv2021

Photoluminescence of ion-synthesized 9R-Si inclusions in SiO2/Si structure: effect of irradiation dose and oxide film thickness

The study of hexagonal silicon polytypes attracts special attention due to their unique physical properties compared to the traditional cubic phase of Si. Thus, for some hexagonal phases, a significant improvement in the emission properties has been demonstrated. In this work, the luminescent properties of SiO2/Si structures irradiated with Kr+ ions at different doses and annealed at 800 °C have been systematically investigated. For such structures, a photoluminescence line at ~ 1240 nm is observed and associated with the formation of hexagonal 9R-Si phase inclusions. It is found that the variation in the thickness of oxide film and the relative position of ion distribution profile and film / substrate interface leads to a regular change in the luminescence intensity. The nature of the observed dependencies is discussed as related to the interplay between the desirable 3C-Si -> 9R-Si structural transition and generation of nonradiative defects. The revealed regularities suggest optimal ion irradiation conditions for synthesis of optically active 9R-Si phase in diamond-like silicon.

cond-mat.mtrl-sci↗

arXiv2021

Ion implantation in \b{eta}-Ga2O3: physics and technology

Gallium oxide and in particular its thermodynamically stable \b{eta}-Ga2O3 phase is within the most exciting materials in research and technology nowadays due to its unique properties, such as an ultra-wide band gap and a very high breakdown electric field, finding a number of applications in electronics and optoelectronics. Ion implantation is a traditional technological method used in these fields, and its well-known advantages can contribute greatly to the rapid development of physics and technology of Ga2O3-based materials and devices. Here, the current status of ion beam implantation in \b{eta}-Ga2O3 is reviewed. The main attention is paid to the results of experimental study of damage under ion irradiation and the properties of Ga2O3 layers doped by ion implantation. The results of ab initio theoretical calculations of the impurities and defects parameters are briefly presented, and the physical principles of a number of analytical methods used to study implanted gallium oxide layers are highlighted. The use of ion implantation in the development of such Ga2O3-based devices as metal oxide field effect transistors, Schottky barrier diodes, and solar-blind UV detectors, is described together with systematical analysis of the achieved values of their characteristics. Finally, the most important challenges to be overcome in this field of science and technology are discussed.

cond-mat.mtrl-sci↗

arXiv2020

Design and Simulation of Memristor-Based Artificial Neural Network for Bidirectional Adaptive Neural Interface

This article proposes a general approach to the simulation and design of a multilayer perceptron (MLP) network on the basis of cross-bar arrays of metal-oxide memristive devices. The proposed approach uses the ANNM theory, tolerance theory, simulation methodology and experiment design. The tolerances analysis and synthesis process is performed for the ANNM hardware implementation on the basis of two arrays of memristive microdevices in the original 16x16 cross-bar topology being a component of bidirectional adaptive neural interface for automatic registration and stimulation of bioelectrical activity of a living neuronal culture used in robotics control system. The ANNM is trained for solving a nonlinear classification problem of stable information characteristics registered in the culture grown on a multi-electrode array. Memristive devices are fabricated on the basis of a newly engineered Au/Ta/ZrO2(Y)/Ta2O5/TiN/Ti multilayer structure, which contains self-organized interface oxide layers, nanocrystals and is specially developed to obtain robust resistive switching with low variation of parameters. An array of memristive devices is mounted into a standard metal-ceramic package and can be easily integrated into the neurointerface circuit. Memristive devices demonstrate bipolar switching of anionic type between the high-resistance state and low-resistance state and can be programmed to set the intermediate resistive states with a desired accuracy. The ANNM tuning, testing and control are implemented by the FPGA-based control subsystem. All developed models and algorithms are implemented as Python-based software.

cs.ET↗