Telegram: @adase_seminar; Twitter: @ADASEgroup
Autumn Session

Date

Speakers

Materials

04 October, 16-30 -- 17-30
Skoltech, Red Building, 359

Topic: Regret bounds of Bayesian optimization

Presenter: Alexey Zaytsev

Abstract: We consider proofs for regret bounds in Bayesian optimization. We'll start with upper bounds and try to continue with the recent (2018) lower bounds looking through some useful techniques and facts along the way.

Paper 1, Paper 2, Paper 3

26 September, 14 -- 16

Skoltech, Red Building, 335

Invited talk: Safer reinforcement learning for robotics,
Presenter: Ivan Koryakovskiy (Delft Tech. Univ.)


This talk will discuss the influence of different prior knowledge on the robot damage-risk and learning performance, where prior knowledge ranges from physics-based assumptions, such as the robot construction and material properties, to the knowledge of the task curriculum, or the approximate model possibly coupled with a nominal controller. Among several conclusions, the provided results reveal the importance of having the approximate model for increasing safety during learning.

21 September, 15 -- 16
Skoltech, Red Building, 333

Dmitrii Smoliakov

  • Bayesian SVM Model

[by Dmitrii]

07 September, 17 -- 20

Skoltech, Red Building, 335

Oleg Voinov

  •  When Recurrent Models Don't Need to be Recurrent  John Miller and Moritz Hardt  [paper]

[by Oleg]

Summer Session

Date

Speakers

Materials

ICML’18 Non-Convex Optimization Workshop

27 July, 17 -- 20
Skoltech, Red Building, 335

Ivan Nazarov

  • Incremental Consensus based Collaborative Deep Learning. Zhanhong Jiang [paper] [sup]
  • Block Mean Approximation for Efficient Second Order Optimization. Yao Lu  [paper] [sup]

Dmitrii Smoliakov

[by Ivan 1, 2]

[by Dmitrii]

ICML’18 Non-Convex Optimization Workshop

3 August, 17 -- 20
Skoltech, Red Building, 335

Evgenii Egorov

  • Using Mode Connectivity for Loss Landscape Analysis. Akhilesh Gotmare, Nitish Shirish Keskar, Caiming Xiong, Richard Socher. [paper] [sup]
  • Loss Surfaces, Mode Connectivity, and Fast Ensembling of DNNs Timur Garipov et al. [paper]

Ivan Nazarov

  • Differential Dynamic Programming for Structured Prediction and Attention Arthur Mensch, Mathieu Blondel  [paper]

[by Evgenii], [video]

[by Ivan], [video]

ICML’18 Non-Convex Optimization Workshop

10 August, 16 -- 20
Skoltech, Red Building, 335

Dmitrii Smoliakov

  • Robust Learning of Trimmed Estimators via Manifold Sampling. Matt Menickelly, Stefan Wild. [paper] [sup]

[by Dmitrii], [video]

ICML’18 Non-Convex Optimization Workshop

17 August, 17 -- 20
Skoltech, Red Building,  335

Evgenii Egorov

  • Entropy-SGD: Biasing Gradient Descent Into Wide Valleys Pratik Chaudhari [paper]

Ivan Nazarov

  • OptNet: Differentiable Optimization as a Layer in Neural Networks Brandon Amos [paper]

[by Evgenii]

[by Ivan]

RNN & CNN

24 August, 17 -- 20
Skoltech, Red Building,  335

Denis Volkhonskiy

  • Attention is all your need (Review)

[by Denis]

RNN & CNN

31 August, 17 -- 20
Skoltech, Red Building,  335

Vage Egiazarian

  • WaveNet (Review)

[by Vage pdf, pptx]

TBA