cv

PhD Student

University of Amsterdam

Reasoning, planning and code generation with applications to embodied agents.

Research Assistant

Artificial Intelligence Research Institute

Topological Deep Learning and its applications to Natural Language Processing.

• Developed a representation learning method that captures the topology of the data space.
• Investigated properties of Transformer attention mechanism through topological data analysis (e.g. persistent homology).

Research Intern

Huawei Technologies

Adversarial text generation aimed at evading AI classifiers.

• Developed a topological loss aimed at generating adversarial examples for text classifiers.

Research Assistant

DeepPavlov.ai

Large-scale intent classification.

• Participated in Amazon Alexa Prize SocialBot Challenge and created a module of flexible large-scale intent classification system.
• During a ``READABLE'' competition, aimed at creating an automatic essay review system for state exams, designed modules for structure extraction and topic modeling.

Master of Science

Skolkovo Institute of Science and Technology (Skoltech)

Artificial Intelligence

• Deep Learning
• Reinforcement Learning
• Natural Language Processing
• Computer Vision
• Optimization

Bachelor of Science

Moscow Institute of Physics and Technology (MIPT)

Physics and Applied Mathematics

• Theoretical Physics
• Discrete Mathematics

Intrinsic dimension estimation for robust detection of ai-generated texts

NeurIPS 2023

Topological intrinsic dimension of the space of AI-generated texts and natural texts in various languages differ enough to be used as a robust feature for detecting AI-generated texts.

STREAM: Embodied Reasoning through Code Generation

Multi-modal Foundation Model meets Embodied AI Workshop @ ICML 2024

A modular approach based on code generation to ground LLM-based planners in the environmental context and enable reasoning about past experiences.

Learning topology-preserving data representations

ICLR 2023

A method for learning topology-preserving data representations that can be used for dimension reduction and visualization.

Acceptability Judgements via Examining the Topology of Attention Maps

EMNLP 2022

A special topological discrepancy measure between attention maps of correct and incorrect sentences can be used as a robust feature for binary grammatical acceptability judgments.

Artificial Text Detection via Examining the Topology of Attention Maps

EMNLP 2021

A robutst method for detecting AI-generated texts based on the topological features of attention maps.