RTEmagicC_people_sasha_02.jpg.jpg

Alexander (Oleksandr) Shekhovtsov, PhD

Assistant Professor
Czech Technical University in Prague
Faculty of Electrical Engineering
Department of Cybernetics

Karlovo namesti 13, 121 35 Prague 2
Czech Republic

shekhole@fel.cvut.cz

Biography:

Graduated in applied mathematics from the National Technical University of Ukraine “Kiev Polytechnic Institute” in 2005. In 2014 received PhD from Czech Technical University in Prague, specialization applied mathematics. In 2014-2017 postdoctoral researcher at Graz University of Technology. Since 2017 researcher at Czech Technical University in Prague.

Research interests:

Currently statistical methods in machine learning and neural networks: dealing with noises and uncertainty, stochastic relaxation, stochastic training of binary and quantized neural networks, Bayesian learning methods. Previously worked in discrete optimization: cuts, flows, matching, assignment, dynamic programming, linear programming, LP relaxation approaches, partial optimality, parallel algorithms, distance transforms.

Recent Talks:

9.12.2020 “Explainable Training of Binary Neural Networks”

Teaching:

Co-lecturer / instructor: “Deep Learning

Instructor: “Patten Recognition

Publications:

  • D. Schlezinger, A. Shekhovtsov, B. Flach (2022). VAE Approximation Error: ELBO and Exponential Families.(ICLR accepted) [openreview]

 

  • A. Shekhovtsov (2021): Bias-variance tradeoffs in single-sample binary gradient estimators. (GCPR) [springer]

 

  • A. Shekhovtsov, V. Yanush (2021): Reintroducing straight-through estimators as principled methods for stochastic binary networks. (GCPR) [springer]

 

  • A. Livocka, A. Shekhovtsov (2021). Initialization and Transfer Learning of Stochastic Binary Networks From Real-Valued Ones. BiVision @ CVPR. [details]

 

  • A. Shekhovtsov, V. Yanush, B. Flach (2020). Path Sample-Analytic Gradient Estimators for Stochastic Binary Networks. NeurIPS. [3 min talk] [details]

 

  • S. Tourani, A. Shekhovtsov, C. Rother, B. Savchynskyy (2020). Taxonomy of dual block-coordinate ascent methods for discrete energy minimization. AISTATS. [details]

 

  • P. Knobelreiter, C. Sormann, A. Shekhovtsov, F. Fraundorfer, T. Pock (2020). Belief propagation reloaded: Learning bp-layers for labeling problems. CVPR. [details]

 

  • Shekhovtsov, A. and Flach, B. (2019). Feed-forward Propagation in Probabilistic Neural Networks with Categorical and Max Layers, ICLR, [paper] [bib]

  • Shekhovtsov, A. and Flach, B. (2018). Stochastic Normalizations as Bayesian Learning, ACCV, [arXiv] [bib]

  • Shekhovtsov, A. and Flach, B. (2018). Normalization of neural networks using analytic variance propagation. [arXiv] [bib]
     
  • Tourani, S.; Shekhovtsov, O.; Rother, C.; Savchynskyy, B. (2018). MPLP++: Fast, Parallel Dual Block-Coordinate Ascent for Dense Graphical Models, ECCV, [www] [bib]

 

  • Shekhovtsov, A., Flach, B., and Busta, M. (2018). Feed-forward uncertainty propagation in belief and neural networks. [arXiv] [bib]

  • Flach, B., Shekhovtsov, A., and Fikar, O. (2017). Generative learning for deep networks. CoRR, abs/1709.08524 [arXiv] [bib]

  • G. Munda, A. Shekhovtsov, P. Knöbelreiter, T. Pock (2017): Scalable Full Flow with Learned Binary Descriptors, GCPR, [arXiv], [bib]

  • A. Shekhovtsov, P. Swoboda and B. Savchynskyy (2017): Maximum Persistency via Iterative Relaxed Inference with Graphical Models, PAMI [preprint] [bib] [code]

 

  • Knöbelreiter, P., Reinbacher, C., Shekhovtsov, A., and Pock, T. (2017). End-to-end training of hybrid CNN-CRF models for stereo. CVPR, to appear. [arXiv] [bib]

  • Li, M., Shekhovtsov, A., and Huber, D. (2016). Complexity of discrete energy minimization problems, ECCV. [arXiv] [spotlight] [bib]

  • Kirillov, A., Shekhovtsov, A., Rother, C., and Savchynskyy, B. (2016). Joint m-best-diverse labelings as a parametric submodular minimization. NIPS, [arXiv] [bib]

 

  • A. Shekhovtsov, C. Reinbacher, G. Graber and T. Pock: Solving Dense Image Matching in Real-Time using Discrete-Continuous Optimization, CVWW 2016 [pdf], [bib]

  • A. Shekhovtsov, P. Swoboda and B. Savchynskyy: Maximum Persistency via Iterative Relaxed Inference with Graphical Models, CVPR 2015 [pdf], [bib], [slides], [poster], [code].

  • A. Shekhovtsov: Higher Order Maximum Persistency and Comparison Theorems, CVIU (SI on Inference & Learning of Graphical Models) 2014 [preprint].

  • A. Shekhovtsov: Maximum Persistency in Energy Minimization, CVPR 2014 [pdf], [bib], [slides]. Research Report [arXiv], [pdf], [bib].

  • A. Shekhovtsov: Exact and Partial Energy Minimization in Computer Vision, PhD Thesis, February 2013. [pdf], [bib], [slides].

  • A. Shekhovtsov, P. Kohli and C. Rother: Curvature Prior for MRF-Based Segmentation and Shape Inpainting, DAGM 2012. [pdf],[bib], [slides].

  • A. Shekhovtsov, V. Hlavac: A Distributed Mincut/Maxflow Algorithm Combining Path Augmentation and Push-Relabel, IJCV 2012. [pdf], [bib], [slides], [code], [preprint].

  • A. Shekhovtsov, V. Hlavac: On Partial Opimality by Auxiliary Submodular Problems, Control Systems and Computers, 2011(2). [pdf], [bib].

  • A. Shekhovtsov, V. Hlavac: Joint Image GMM and Shading MAP Estimation, ICPR 2010. [pdf], [bib], Technical Report [pdf]

  • A. Shekhovtsov, V. Hlavac: A Lower Bound by One-against-all Decomposition for Potts Model Energy Minimization, CVVW 2008. [pdf], [bib], [slides]
     
  • P. Kohli, A. Shekhovtsov, C. Rother, V. Kolmogorov, P. Torr: On partial optimality in multi-label MRFs. ICML 2008: Proceedings of the 25th International Conference on Machine Learning. [pdf], [bib], [slides]

  • Shekhovtsov, J.D. Garcia-Arteaga, T. Werner: A Discrete Search Method for Multi-modal Non-Rigid Image Registration. NORDIA 2008: Proceedings of the 2008 IEEE CVPR Workshop on Non-Rigid Shape Analysis and Deformable Image Alignment. [pdf], [bib], [slides]

  • Shekhovtsov, I. Kovtun, V. Hlavac: Efficient MRF Deformation Model for Non-Rigid Image Matching. CVIU 2008. [ScienceDirect]  ([preprint] / [ealier version]), [bib], [videos]

  • T. Werner, A. Shekhovtsov: Unified Framework for Semiring-Based Arc Consistency and Relaxation Labeling. Computer Vision Winter Workshop , St. Lambrecht, Austria, February 2007.

  • Shekhovtsov: Supermodular decomposition of structural labeling problem (in Russian). Control Systems and Computers #1, 2006, pp.39-48; Kiev, Ukraine. [pdf], [bib]

  • B. Flach, D. Schlesinger, A. Shekhovtsov: A Higher Order MRF-Model for Stereo-Reconstruction, Pattern Recognition, LNCS vol. 3175, 2004, 440-446. [link]

 

Technical Reports

  • A. Shekhovtsov, P. Kohli, C. Rother: Curvature Prior for MRF-based Segmentation and Shape Inpainting, Research Report CTU--CMP--2011--11, Czech Technical University. [pdf], [bib]

 

  • A. Shekhovtsov, V. Hlavac: A Distributed Mincut/Maxflow Algorithm Combining Path Augmentation and Push-Relabel, Research Report K333--43/11, CTU--CMP--2011—03, Czech Technical University. [pdf], [bib]

  • A. Shekhovtsov, V. Hlavac: Joint Image GMM and Shading MAP Estimation, Research Report K333–35/10, CTU–CMP–2010–03, Czech Technical University.
    [pdf], [bib]

 

  • A. Shekhovtsov, V. Kolmogorov, P.Kohli, V. Hlavac, C. Rother, P. Torr: LP-relaxation of binarized energy minimization, Research Report CTU--CMP--2007—27, Czech Technical University, update 2008. [pdf], [bib]

 

 

 

voxelmrf

optimality

bnd2