Anton Osokin

Isomorphic Labs

Research Scientist

preferred contact:

Bio | Publications | Teaching

I'm an ML researcher most excited about working with complicated objects with a lot of structure. Currently, I work with molecules at Isomorphic Labs. In the past, I used to work with code, text and images as an associate professor at HSE University, Moscow, Russia and researcher at Yandex lab/Yandex Research. I did both my undergrad and PhD studies in computer science at Lomonosov Moscow State University with Bayes Group and a postdoc in ML/CV at INRIA/École Normale Supérieure in Paris.

Selected publications

SPARQLing Database Queries from Intermediate Question Decompositions
Irina Saparina and Anton Osokin
In proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2021
[paper and supplementary] [bibtex] [code and data] 
@inproceedings{saparina21sparqling,
    title = {{SPARQLing} Database Queries from Intermediate Question Decompositions},
    author = {Irina Saparina and Anton Osokin},
    booktitle = {proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
    year = {2021} }
OS2D: One-Stage One-Shot Object Detection by Matching Anchor Features
Anton Osokin, Denis Sumin and Vasily Lomakin
In proceedings of the European Conference on Computer Vision (ECCV), 2020
[paper and supplementary] [bibtex] [code and data] [teaser video (1 min)] [video (10 min)] 
@inproceedings{osokin20os2d,
    title = {{OS2D}: One-Stage One-Shot Object Detection by Matching Anchor Features},
    author = {Anton Osokin and Denis Sumin and Vasily Lomakin},
    booktitle = {proceedings of the European Conference on Computer Vision (ECCV)},
    year = {2020} }
Quantifying Learning Guarantees for Convex but Inconsistent Surrogates
Kirill Struminsky, Simon Lacoste-Julien and Anton Osokin
Advances in Neural Information Processing Systems (NeurIPS), 2018
[paper and supplementary] [bibtex] [teaser video (3 min)] [seminar talk (90 min)] 
@inproceedings{struminsky18consistency,
    title = {Quantifying Learning Guarantees for Convex but Inconsistent Surrogates},
    author = {Kirill Struminsky and Simon Lacoste-Julien and Anton Osokin},
    booktitle = {Advances in Neural Information Processing Systems (NIPS)},
    year = {2018} }
Marginal Weighted Maximum Log-likelihood for Efficient Learning of Perturb-and-Map Models
Tatiana Shpakova, Francis Bach, Anton Osokin
In proceedings of the international conference on Uncertainty in Artificial Intelligence (UAI), 2018
[paper] [bibtex] 
@inproceedings{shpakova2018,
    author    = {Tatiana Shpakova and Francis Bach and Anton Osokin},
    title     = {Marginal Weighted Maximum Log-likelihood for Efficient Learning of Perturb-and-Map Models},
    booktitle = {UAI},
    year      = {2018} }
SEARNN: Training RNNs with Global-Local Losses
Rémi Leblond, Jean-Baptiste Alayrac, Anton Osokin, Simon Lacoste-Julien
In proceedings of the International Conference on Learning Representations (ICLR), 2018
[paper and supplementary] [bibtex] [webpage/data] [code] 
@inproceedings{searnn2018leblond,
    author    = {Leblond, R\'emi and Alayrac, Jean-Baptiste and Osokin, Anton and Lacoste-Julien, Simon},
    title     = {\textsc{SeaRnn}: Training RNNs with Global-Local Losses},
    booktitle = {ICLR},
    year      = {2018} }
On Structured Prediction Theory with Calibrated Convex Surrogate Losses
Anton Osokin, Francis Bach and Simon Lacoste-Julien
Advances in Neural Information Processing Systems (NIPS) (oral presentation), 2017
[paper and supplementary] [bibtex] [code] [poster] [teaser video (3 min)] [NIPS talk (15 min)] 
@inproceedings{osokin17consistency,
    title = {On Structured Prediction Theory with Calibrated Convex Surrogate Losses},
    author = {Anton Osokin and Francis Bach and Simon Lacoste-Julien},
    booktitle = {Advances in Neural Information Processing Systems (NIPS)},
    year = {2017} }
GANs for Biological Image Synthesis
Anton Osokin, Anatole Chessel, Rafael E. Carazo Salas and Federico Vaggi
In proceedings of the International Conference on Computer Vision (ICCV), 2017
[paper and supplementary] [bibtex] [code/data] [poster] [ICCV spotlight (3 min)] 
@InProceedings{osokin2017biogans,
    author = {Anton Osokin and Anatole Chessel and Rafael E. Carazo Salas and Federico Vaggi},
    title = {{GANs} for Biological Image Synthesis},
    booktitle = {proceedings of the International Conference on Computer Vision (ICCV)},
    year = {2017} }
Minding the Gaps for Block Frank-Wolfe Optimization of Structured SVMs
Anton Osokin*, Jean-Baptiste Alayrac*, Isabella Lukasewitz, Puneet K. Dokania, and Simon Lacoste-Julien
*equal contribution
In proceedings of the International Conference on Machine Learning (ICML), 2016
[paper and supplementary] [ICML talk (15 min)] [bibtex] [webpage] [code] 
@InProceedings{osokin16gapBCFW,
     author = {Anton Osokin and Jean-Baptiste Alayrac and Isabella Lukasewitz and Puneet K. Dokania and Simon Lacoste-Julien},
     title = {Minding the Gaps for Block {F}rank-{W}olfe Optimization of Structured {SVM}s},
     booktitle = {proceedings of the International Conference of Machine Learning (ICML)},
     year = {2016} }
Breaking Sticks and Ambiguities with Adaptive Skip-gram
Sergey Bartunov, Dmitry Kondrashkin, Anton Osokin, and Dmitry Vetrov
In proceedings of the International Conference on Artificial Intelligence and Statistics (AISTATS), 2016
[paper] [supplementary] [bibtex] [code] 
@inproceedings{bartunov16adagram,
    title = {Breaking Sticks and Ambiguities with Adaptive {S}kip-gram},
    author = {Sergey Bartunov and Dmitry Kondrashkin and Anton Osokin and Dmitry Vetrov},
    booktitle = {proceedings of the International Conference on Artificial Intelligence and Statistics (AISTATS)},
    year = {2016} }
Context-aware CNNs for person head detection
Tuan-Hung Vu, Anton Osokin, and Ivan Laptev
In proceedings of the International Conference on Computer Vision (ICCV), 2015
[paper] [webpage] [bibtex] [code] 
@inproceedings{vu15contextCNN,
    title = {Context-aware {CNNs} for person head detection},
    author = {Tuan-Hung Vu and Anton Osokin and Ivan Laptev},
    booktitle = {proceedings of the International Conference on Computer Vision (ICCV)},
    year = {2015} }
Tensorizing Neural Networks
Alexander Novikov, Dmitry Podoprikhin, Anton Osokin, and Dmitry Vetrov
Advances in Neural Information Processing Systems (NIPS), 2015
[paper] [bibtex] [code] 
@inproceedings{novikov15tensornet,
    title = {Tensorizing Neural Networks},
    author = {Alexander Novikov and Dmitry Podoprikhin and Anton Osokin and Dmitry Vetrov},
    booktitle = {Advances in Neural Information Processing Systems (NIPS)},
    year = {2015} }
Submodular relaxation for inference in Markov random fields
Anton Osokin and Dmitry Vetrov
IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 37(7): 1347-1359, 2015
[paper] [bibtex] [code] 
@article{osokin15smr,
    Title = {Submodular relaxation for inference in {M}arkov random fields},
    Author = {Anton Osokin and Dmitry Vetrov},
    Journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)},
    Year = {2015},
    Number = {7},
    Pages = {1347--1359},
    Volume = {37} }
Perceptually Inspired Layout-aware Losses for Image Segmentation
Anton Osokin and Pushmeet Kohli
In proceedings of the European Conference on Computer Vision (ECCV), 2014
[paper] [bibtex] [code] 
@inproceedings{osokin14losses,
    title = {Perceptually Inspired Layout-aware Losses for Image Segmentation},
    author = {Anton Osokin and Pushmeet Kohli},
    booktitle = {proceedings of the European Conference on Computer Vision (ECCV)},
    year = {2014} }
Putting MRFs on a Tensor Train
Alexander Novikov, Anton Rodomanov, Anton Osokin, and Dmitry Vetrov
In proceedings of the International Conference on Machine Learning (ICML), 2014
[paper] [supplementary] [bibtex] [code] 
@inproceedings{novikov14tensorTrainMRF,
    title = {Putting {MRFs} on a Tensor Train},
    author = {Alexander Novikov and Anton Rodomanov and Anton Osokin and Dmitry Vetrov},
    booktitle = {proceedings of the International Conference on Machine Learning (ICML)},
    year = {2014} }
A Principled Deep Random Field Model for Image Segmentation
Pushmeet Kohli, Anton Osokin, and Stefanie Jegelka
In proceedings of the Computer Vision and Pattern Recognition (CVPR), 2013
[paper] [supplementary] [bibtex] [code] 
@inproceedings{kohli13segmentation,
    title = {A Principled Deep Random Field Model for Image Segmentation},
    author = {Pushmeet Kohli and Anton Osokin and Stefanie Jegelka},
    booktitle = {proceedings of the Computer Vision and Pattern Recognition (CVPR)},
    year = {2013} }
Minimizing Sparse High-Order Energies by Submodular Vertex-Cover
Andrew Delong, Olga Veksler, Anton Osokin, and Yuri Boykov
Advances in Neural Information Processing Systems (NIPS), 2012
[paper] [bibtex] 
@inproceedings{delong12vertexCover,
    title = {Minimizing Sparse High-Order Energies by Submodular Vertex-Cover},
    author = {Andrew Delong and Olga Veksler and Anton Osokin and Yuri Boykov},
    booktitle = {Advances in Neural Information Processing Systems (NIPS)},
    year = {2012} }
Fast Approximate Energy Minimization with Label Costs
Andrew Delong*, Anton Osokin*, Hossam Isack, Yuri Boykov
*equal contribution
International Journal of Computer Vision (IJCV), 96(1):1-27, 2012
[paper] [bibtex] [code] 
@article{delong12labelCosts,
    Title = {Fast Approximate Energy Minimization with Label Costs},
    Author = {Andrew Delong and Anton Osokin and Hossam Isack and Yuri Boykov},
    Journal = {International Journal of Computer Vision (IJCV)},
    Year = {2012},
    Number = {1},
    Pages  = {1--27},
    Volume  = {96} }
Submodular Decomposition Framework for Inference in Associative Markov Networks with Global Constraints
Anton Osokin, Dmitry Vetrov, and Vladimir Kolmogorov
In proceedings of the Computer Vision and Pattern Recognition (CVPR), 2011
[paper] [techreport] [bibtex] [code] 
@inproceedings{osokin11smd,
    title = {Submodular Decomposition Framework for Inference in Associative {M}arkov Networks with Global Constraints},
    author = {Anton Osokin and Dmitry Vetrov and Vladimir Kolmogorov},
    booktitle = {proceedings of the Computer Vision and Pattern Recognition (CVPR)},
    year = {2011} }
Fast Approximate Energy Minimization with Label Costs
Andrew Delong*, Anton Osokin*, Hossam Isack, and Yuri Boykov
*equal contribution
In proceedings of the Computer Vision and Pattern Recognition (CVPR), 2010
[paper] [bibtex] [code] 
@inproceedings{delong10labelCosts,
    title = {Fast Approximate Energy Minimization with Label Costs},
    author = {Andrew Delong and Anton Osokin and Hossam Isack and Yuri Boykov},
    booktitle = {proceedings of the Computer Vision and Pattern Recognition (CVPR)},
    year = {2010} }

PhD thesis

Submodular relaxation for energy minimization in Markov random fields
Anton Osokin
Lomonosov Moscow State University. 2014. In Russian
[text (pdf) (in Russian)] [synopsis (pdf) (in Russian)] [code]
see TPAMI 2015 paper for the English version

Teaching

Deep learning at CS HSE, Moscow, Russia. Lecturer. 2018, 2019-spring, 2019-fall, 2020, 2021
All the materials (in Russian) are available online (lecture slides, seminars, recorded lectures)
Introduction to discrete optimization at CentraleSupélec, Châtenay-Malabry, France. Co-lecturer with Karteek Alahari. 2016
Graphical models at CMC MSU. Seminars and practical sessions. Lecturers: Dmitry Vetrov, Dmitry Kropotov.
2012, 2013, 2014
Graphical models at Yandex Data Analysis School. Seminars and practical sessions. Lecturers: Victor Lempitsky, Dmitry Vetrov.
2011, 2012, 2013
Machine Learning at CMC MSU. Practical sessions.
2012, 2013, 2014
Scientific seminar on Bayesian methods of machine learning at CMC MSU. Co-organizer together with Dmitry Vetrov and Dmitry Kropotov.
2010-2014