Taras Khakhulin

Graduated from: Skolkovo Institute of Science and Technology

Research interests: Deep Learning, Image Synthesis, Adversarial Learning

Thesis title: Deep learning models for three-dimensional scenes

Publications:

Khakhulin, T., Korzhenkov, D., Solovev, P., Ardelean T., Sterkin, G ., Lempitsky, V. (2022). Stereo Magnification with Multi-Layer Images  2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),

Anokhin, I., Demochkin, K., Khakhulin, T., Sterkin, G ., Lempitsky, V., Korzhenkov, D. (2021). Image generators with conditionally-independent pixel synthesis. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),

Anokhin, I., Solovev, P., Korzhenkov, D., Kharlamov, A., Khakhulin, T., Silvestrov, A., Nikolenko, S., Lempitsky, V., & Sterkin, G. (2020). High-Resolution Daytime Translation Without Domain Labels. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 7485–7494,

Simple heuristics for efficient parallel tensor contraction and quantum circuit simulation
Roman Schutski, Taras Khakhulin, Ivan Oseledets, and Dmitry Kolmakov, Phys. Rev. A 102, 062614 – Published 28 December 2020

Workshops:

Learning Elimination Ordering for Tree Decomposition Problem, Workshop “Learning Meets Combinatorial Algorithms” at NeurIPS2020

Robust word vectors: context-informed embeddings for noisy texts, Workshop “Noisy User-generated Text” at EMNLP 2018

Short biography:

Taras Khakhulin graduated from the Moscow Institute of Physics and Technology in 2018 with a Bachelor of Science degree and from the Skolkovo Institute of Science and Technology in 2020 with a Master of Science (degree with honor). During his undergraduate studies, Taras contributed to the word representation field for texts with misspellings and typos. He also participated in the development open-source library for natural language understanding presented in ACL 2018. During his Master’s studies Taras investigated the application of reinforcement learning for discrete optimization problems, his findings were presented at NeurIPS 2020 workshop. 

Currently, Taras is a PhD student at Skolkovo Institute of Science and Technology, where he is a part of the Computer Vision team under the supervision of Victor Lemptsky. His findings have shown the possibility of creating image generators without spatial convolutions. The main interest of Taras’s research is the study of representations for novel view synthesis. He described scene-adapted representation that allows novel view synthesis on devices and developed methods to generate such representation without optimization procedure from the arbitrary number of views. Most of the research was sponsored and done at the Samsung AI Center Moscow and presented at CVPR. At the moment, research is devoted to one-shot human photography animation and head reconstruction.