HSE-Yandex autumn school on generative models

26 Nov 2019, 09:00 → 29 Nov 2019, 19:00 Europe/Moscow

room "Oxford" (Yandex, Moscow)

Ivan Arzhantsev (HSE), Alexey Naumov (HSE), Andrey Ustyuzhanin (HSE), Dmitriy Vetrov (HSE)

HSE University and Yandex invite you to joint autumn school on generative models aimed at undergraduate/graduate students and young postdoctoral fellows from pure and applied mathematics. This intense four-day workshop will consist of 3 interdisciplinary mini-courses, list of invited talks, poster session by the participants and master-classes by industrial partners. Key topics of this school:

• Generative Adversarial Nets
• Statistical and Computational Optimal Transport
• Bayesian methods in machine learning

This school is supported by the RSF grant N19-71-30020 "Applications of probabilistic artificial neural generative models to development of digital twin technology for Non-linear stochastic systems" and organised by three laboratories of HSE University:

• LAMBDA (Laboratory of Methods for Big Data Analysis)
• DeepBayes (Centre of Deep Learning and Bayesian Methods)
• HDI Lab (International laboratory of stochastic algorithms and high-dimensional inference)

Yandex is an industry partner of the school.

Mini-courses:

• Optimization methods for optimal transport by Pavel Dvurechensky (WIAS Berlin)
• Generative models by Denis Derkach (HSE University)
• Introduction to scalable Bayesian methods by Dmitry Vetrov, Ekaterina Lobacheva (HSE University) and Nadia Chirkova (HSE University)

Invited talks:

• Artem Babenko (Yandex, HSE)
• Evgeny Burnaev (Skoltech)
• Viktor Lempitsky (Skoltech)
• Ivan Oseledets (Skoltech)
• Maxim Panov (Skoltech, HSE)
• Vladimir Spokoiny (WIAS, HSE)
• TBC

Poster Submission Guidelines:

Only posters of submitted and accepted abstracts will be offered presentation. Please note the following information for the preparation of your poster. Please bring your printed poster with you to the Conference.

Poster Preparation

• The poster layout is PORTRAIT.
• Please prepare your poster to fit the dimensions below. The poster can be prepared either on one sheet or few sheets of paper.
• The dimensions of the poster should not exceed 59.4 cm wide x 84 cm long (23.4 inches wide x 33.1 inches long). The recommended but not mandatory format is A1.
• Allocate the top of the poster for the title and authors as stated on the submitted abstract.
• Pins will be available for the mounting of posters.

Derkach_1 lecture.pdf

Dvurechensky lectures.pdf

HSE University and Yandex.pdf

Tuesday, 26 November

09:45 → 10:00 Organization: Welcome

10:00 → 11:00 LAMBDA: Simple generative models. Generative adversarial networks 1

LAMBDA talks

Convener: Prof. Denis Derkach (National Research University Higher School of Economics (RU))

Derkach_1 lecture.pdf

Seminar

11:00 → 11:30 Coffee Break

11:30 → 12:30 LAMBDA: Simple generative models. Generative adversarial networks 2

LAMBDA talks

Convener: Prof. Denis Derkach (National Research University Higher School of Economics (RU))

Derkach_1 lecture.pdf

Seminar

12:30 → 14:00 Lunch

14:00 → 15:00 HDI Lab: Optimization methods for optimal transport 1

In this lecture I will give a short introduction to optimal transport problem with some motivating examples from modern machine learning, including image retrieval and classification. The effectiveness of optimal transport in applications comes with the price of heavy computations and I will discuss two types of methods which allow to efficiently compute optimal transport distance (Wasserstein distance). The first method is Sinkhorn's algorithm and the second one is accelerated gradient method. If time allows I will also discuss a next level problem of finding Wasserstein barycenter of a set of measures, which works quite well in image analysis. Numerical methods for this problem will be discussed.

Convener: Dr Pavel Dvurechensky (WIAS Berlin)

Dvurechensky lectures.pdf

15:00 → 15:30 Tea Break

15:30 → 16:30 HDI Lab: Optimization methods for optimal transport 2

In this lecture I will give a short introduction to optimal transport problem with some motivating examples from modern machine learning, including image retrieval and classification. The effectiveness of optimal transport in applications comes with the price of heavy computations and I will discuss two types of methods which allow to efficiently compute optimal transport distance (Wasserstein distance). The first method is Sinkhorn's algorithm and the second one is accelerated gradient method. If time allows I will also discuss a next level problem of finding Wasserstein barycenter of a set of measures, which works quite well in image analysis. Numerical methods for this problem will be discussed.

Convener: Dr Pavel Dvurechensky (WIAS Berlin)

Dvurechensky lectures.pdf

16:30 → 17:30 Talks: New directions in generative models and high-dimensional MC methods

Convener: Prof. Eric Moulines (Ecole Polytechnique, HSE University)

Moulines_Yandex_2019.pdf

Wednesday, 27 November

10:00 → 11:00 HDI Lab: Computational and statistical optimal transport 3

In this lecture I will give a short introduction to optimal transport problem with some motivating examples from modern machine learning, including image retrieval and classification. The effectiveness of optimal transport in applications comes with the price of heavy computations and I will discuss two types of methods which allow to efficiently compute optimal transport distance (Wasserstein distance). The first method is Sinkhorn's algorithm and the second one is accelerated gradient method. If time allows I will also discuss a next level problem of finding Wasserstein barycenter of a set of measures, which works quite well in image analysis. Numerical methods for this problem will be discussed.

Convener: Dr Pavel Dvurechensky (WIAS Berlin)

Dvurechensky lectures.pdf

11:00 → 11:30 Coffee Break

11:30 → 12:30 HDI Lab: Computational and statistical optimal transport 4

In this lecture I will give a short introduction to optimal transport problem with some motivating examples from modern machine learning, including image retrieval and classification. The effectiveness of optimal transport in applications comes with the price of heavy computations and I will discuss two types of methods which allow to efficiently compute optimal transport distance (Wasserstein distance). The first method is Sinkhorn's algorithm and the second one is accelerated gradient method. If time allows I will also discuss a next level problem of finding Wasserstein barycenter of a set of measures, which works quite well in image analysis. Numerical methods for this problem will be discussed.

Convener: Dr Pavel Dvurechensky (WIAS Berlin)

Dvurechensky lectures.pdf

12:30 → 12:40 Organization: Group Photo 1

12:40 → 14:00 Lunch

14:00 → 15:00 DeepBayes: Introduction to scalable Bayesian methods 1

Conveners: Prof. Dmitry Vetrov (HSE University), Dr Ekaterina Lobacheva (HSE University), Dr Nadia Chirkova (HSE)

assignment_practice.ipynb

assignment_practice_solution.ipynb

bayesian_methods_lecture.pdf

bayesian_methods_problem_set.pdf

bayesian_methods_seminar.pdf

bnn_tutorial.pdf

ysda_bnn_tutorial.pdf

15:00 → 15:30 Tea Break

15:30 → 16:30 DeepBayes: Introduction to scalable Bayesian methods 2

Conveners: Prof. Dmitry Vetrov (HSE University), Dr Ekaterina Lobacheva (HSE University), Dr Nadia Chirkova (HSE)

assignment_practice.ipynb

assignment_practice_solution.ipynb

bayesian_methods_lecture.pdf

bayesian_methods_problem_set.pdf

bayesian_methods_seminar.pdf

bnn_tutorial.pdf

ysda_bnn_tutorial.pdf

Thursday, 28 November

10:00 → 11:00 LAMBDA: Advanced generative models. Introduction to normalizing flows 1

LAMBDA talks

Convener: Prof. Denis Derkach (National Research University Higher School of Economics (RU))

Derkach_1 lecture.pdf

Seminar

11:00 → 11:30 Coffee Break

11:30 → 12:30 LAMBDA: Advanced generative models. Introduction to normalizing flows 2

LAMBDA talks

Convener: Prof. Denis Derkach (National Research University Higher School of Economics (RU))

Derkach_1 lecture.pdf

Seminar

12:30 → 14:00 Lunch

14:00 → 14:30 Yandex Master class: RPGAN: random paths as a latent space for GAN interpretability

14:30 → 15:00 Yandex Master class: Sequence modeling with unconstrained generation order

15:00 → 15:30 Tea Break

15:30 → 16:30 DeepBayes: Introduction to scalable Bayesian methods 3

Conveners: Prof. Dmitry Vetrov (HSE University), Dr Ekaterina Lobacheva (HSE University), Dr Nadia Chirkova (HSE)

assignment_practice.ipynb

assignment_practice_solution.ipynb

bayesian_methods_lecture.pdf

bayesian_methods_problem_set.pdf

bayesian_methods_seminar.pdf

bnn_tutorial.pdf

ysda_bnn_tutorial.pdf

16:30 → 17:30 DeepBayes: Introduction to scalable Bayesian methods 4

Conveners: Prof. Dmitry Vetrov (HSE University), Dr Ekaterina Lobacheva (HSE University), Dr Nadia Chirkova (HSE)

assignment_practice.ipynb

assignment_practice_solution.ipynb

bayesian_methods_lecture.pdf

bayesian_methods_problem_set.pdf

bayesian_methods_seminar.pdf

bnn_tutorial.pdf

ysda_bnn_tutorial.pdf

19:00 → 21:00 Evening session: Conference dinner (В лофте Studiohall. Адрес: Ленинский пр. Д. 49)

Map

Friday, 29 November

10:00 → 10:50 Talks: Bayesian inference vs stochastic optimization

Convener: Prof. Vladimir Spokoiny (WIAS, IITP, HSE)

Moulines_Yandex_2019.pdf

10:50 → 11:30 Talks: Wasserstein-2 Generative Networks

Convener: Prof. Evgeny Burnaev (Skoltech)

Moulines_Yandex_2019.pdf

11:30 → 11:40 Organization: Group Photo 2

11:40 → 12:10 Coffee Break

12:10 → 12:50 Talks: Towards Photorealistic Neural Avatars

Convener: Prof. Viktor Lempitsky (Samsung AI, Skoltech)

Moulines_Yandex_2019.pdf

12:50 → 13:10 Talks: Adaptive Divergence for Rapid Adversarial Optimization

Convener: Maxim Borisyak (Yandex School of Data Analysis (RU))

Moulines_Yandex_2019.pdf

13:10 → 13:30 Talks: (1+ε)-class Classification: an Anomaly Detection Method for Highly Imbalanced or Incomplete Data Sets

Convener: Andrey Ustyuzhanin (Yandex School of Data Analysis (RU))

Moulines_Yandex_2019.pdf

13:30 → 14:40 Lunch

14:40 → 15:20 Talks: Approximation of multivariate functions using deep learning with applications

Convener: Prof. Ivan Oseledets (Skoltech)

Moulines_Yandex_2019.pdf

15:20 → 15:50 Talks: Structure-adaptive manifold estimation

Convener: Mr Nikita Puchkin (HSE)

Moulines_Yandex_2019.pdf

15:50 → 16:20 Tea Break

16:20 → 16:50 Talks: Uncertainty estimation: can your neural network provide confidence for its predictions?

Convener: Prof. Maxim Panov (Skoltech, HSE)

Moulines_Yandex_2019.pdf

16:50 → 17:15 Talks: Fast Simulation Using Generative Adversarial Networks in LHCb

Convener: Mr Artem Maevskiy (HSE University)

Moulines_Yandex_2019.pdf

17:15 → 17:40 Talks: Differentiating the Black-Box: Optimization with Local Generative Surrogates

Convener: Mr Vladislav Belavin (HSE University)

Moulines_Yandex_2019.pdf

17:40 → 17:50 Organization: The end