Albina Klepach

Hi! My research focuses on machine learning, deep learning, and reinforcement learning, with a particular interest in applying these techniques to real-world challenges like learning from raw videos or compiler optimization.

Currently I am focusing on imitation learning solely from video data using latent action models (LAMs). We explored application of object-centric pretraining to enhance the performance of LAMs under distractions. Before, I worked as a Machine Learning Research Engineer at Huawei’s Hisilicon division, where I applied reinforcement learning and graph machine learning to optimize compiler phase ordering and central processor performance.

I hold a Master’s degree in Data Science, where I conducted research on leader-guided evacuation models combining active matter and reinforcement learning under the guidance of Prof. Nikolay Brilliantov. During my Bachelor’s studies in Applied Mathematics and Physics, I explored high-energy astrophysics at the Space Research Institute RAS with Prof. Alexander Lutovinov, focusing on spectral and timing analysis of X-ray pulsars. Additionally, I worked on topological clustering of protein structures during internship at Max Planck MPI-CBG Dresden with Prof. Carl Modes.

In addition to my professional experience, I have worked on various pet projects, including implementing in-context reinforcement learning algorithm, enhancing model-based RL with exploration strategies, and solving combinatorial optimization problems like the Travelling Salesman Problem. Outside of research, I enjoy teaching, watching good films, and practicing acrobatics.

selected publications

ICLR Workshop
Object-Centric Latent Action Learning

Albina Klepach, Alexander Nikulin, and 6 more authors

In ICLR 2025 Workshop on World Models: Understanding, Modelling and Scaling, 2025

Abs Bib HTML

Leveraging vast amounts of unlabeled internet video data for embodied AI is currently bottlenecked by the lack of action labels and the presence of action-correlated visual distractors. Although recent latent action policy optimization (LAPO) has shown promise in inferring proxy-action labels from visual observations, its performance degrades significantly when distractors are present. To address this limitation, we propose a novel object-centric latent action learning framework that focuses on objects rather than pixels. We leverage self-supervised object-centric pretraining to disentangle action-related and distracting dynamics. This allows LAPO to focus on task-relevant interactions, resulting in more robust proxy-action labels, enabling better imitation learning and efficient adaptation of the agent with just a few action-labeled trajectories. We evaluated our method in eight visually complex tasks in the Distracting Control Suite (DCS) and Distracting MetaWorld (DMW). Our results show that object-centric pretraining mitigates the negative effects of distractors by \emph50%, as measured by downstream task performance: average return and success rate (for DCS and DMW, accordingly)
@inproceedings{klepach2025object, title = {Object-Centric Latent Action Learning}, author = {Klepach, Albina and Nikulin, Alexander and Zisman, Ilya and Tarasov, Denis and Derevyagin, Alexander and Polubarov, Andrei and Nikita, Lyubaykin and Kurenkov, Vladislav}, booktitle = {ICLR 2025 Workshop on World Models: Understanding, Modelling and Scaling}, year = {2025}, }
Elsevier
An effective control of large systems of active particles: An application to evacuation problem

Albina Klepach, Egor E. Nuzhin, and 2 more authors

Communications in Nonlinear Science and Numerical Simulation, 2025

Abs Bib HTML Code

Manipulation of large systems of active particles is a serious challenge across diverse domains, including crowd management, control of robotic swarms, and coordinated material transport. The development of advanced control strategies for complex scenarios is hindered, however, by the lack of scalability and robustness of the existing methods, in particular, due to the need of an individual control for each agent. One possible solution involves controlling a system through a leader or a group of leaders, which other agents tend to follow. Using such an approach we develop an effective control strategy for a leader, combining reinforcement learning (RL) with artificial forces acting on the system. To describe the guidance of active particles by a leader we introduce the generalized Vicsek model. This novel method is then applied to the problem of the effective evacuation by a robot-rescuer (leader) of large groups of people from hazardous places. We demonstrate, that while a straightforward application of RL yields suboptimal results, even for advanced architectures, our approach provides a robust and efficient evacuation strategy. The source code supporting this study is publicly available at: https://github.com/cinemere/evacuation.
@article{KLEPACH2025109359, title = {An effective control of large systems of active particles: An application to evacuation problem}, journal = {Communications in Nonlinear Science and Numerical Simulation}, pages = {109359}, year = {2025}, issn = {1007-5704}, doi = {10.1016/j.cnsns.2025.109359}, url = {https://www.sciencedirect.com/science/article/pii/S1007570425007683}, author = {Klepach, Albina and Nuzhin, Egor E. and Tsukanov, Alexey A. and Brilliantov, Nikolay V.}, keywords = {control theory, reinforcement learning, active matter, artificial forces}, }