Математические методы прогнозирования (практика, В.В. Стрижов)/Группа 574, осень 2019

Материал из MachineLearning.

(Различия между версиями)
Перейти к: навигация, поиск
(47 промежуточных версий не показаны.)
Строка 1: Строка 1:
-
{{Main|Численные методы обучения по прецедентам (практика, В.В. Стрижов)}}
+
{{Main|Численные методы обучения по прецедентам (практика, В.В. Стрижов)}}
__NOTOC__
__NOTOC__
-
Short link [ ]
+
Short link [http://bit.ly/IS_B2 bit.ly/IS_B2]
This series of seminars continues the course Bayesian model selection and investigates the theoretical aspects of model selection in various application problems.
This series of seminars continues the course Bayesian model selection and investigates the theoretical aspects of model selection in various application problems.
-
* Seminar 1 (Isachenko)
+
Videolectures are available [https://www.youtube.com/playlist?list=PLk4h7dmY2eYH9RtoKGzxHKji0GLiBzSlZ here].
-
** Plate notation and Bayesian inference in examples
+
 
-
** (reminder Coherent Bayesian Inference)
+
==== Seminar 1 (Isachenko, [[Медиа:Isachenko2019DeepGenerativeModels1.pdf‎|slides]]) ====
-
** Variational inference
+
* Generative models
-
** Variational autoencoder
+
* Applications
-
** ELBO
+
* Autoregressive models (CharRNN, MADE, WaveNet, PixelCNN)
-
* Seminar 2 (Isachenko)
+
 
-
** Analytic methods of approximation
+
==== Seminar 2 (Isachenko, [[Медиа:Isachenko2019DeepGenerativeModels2.pdf‎|slides]]) ====
-
** Statistic sum approximation
+
* Generative vs discriminative
-
** Generative versus discriminative
+
* Latent variable models
-
* Seminar 3 (Isachenko)
+
* Variational Inference
-
** Inference methods of approximation
+
* ELBO
-
** Zoo of variational autoencoders and practical examples
+
* Variational Autoencoder
-
* Seminar 4 (Isachenko)
+
 
 +
==== Seminar 3 (Isachenko, [[Медиа:Isachenko2019DeepGenerativeModels3.pdf‎|slides]]) ====
 +
* Mean field approximation
 +
* Flow models (NICE, RealNVP)
 +
 
 +
==== Seminar 4 (Isachenko, [[Медиа:Isachenko2019DeepGenerativeModels4.pdf‎|slides]]) ====
 +
* VAE Limitations
 +
* Flows in VAE
 +
* Autoregressive flows (MAF, IAF, Parallel WaveNet)
 +
 
 +
==== Seminar 5 (Isachenko, [[Медиа:Isachenko2019DeepGenerativeModels5.pdf‎|slides]]) ====
 +
* IWAE (lower bound, posterior, inactive units)
 +
* ELBO surgery
 +
* VampPrior
 +
 
 +
==== Seminar 6 (Isachenko, [[Медиа:Isachenko2019DeepGenerativeModels6.pdf‎|slides]]) ====
 +
* Autoregressive decoder in VAE
 +
* Posterior collapse, decoder weakening
 +
* Disentangled representations
 +
* beta-VAE
 +
 
 +
==== Seminar 7 (Bakhteev, [https://github.com/bahleg/tex_slides/raw/master/oct_19/slides1_elbo.pdf slides]) ====
 +
* Model selection statement
 +
* ELBO for model selection
 +
* Early Stopping is Nonparametric Variational Inference
 +
* Langevin dynamics
 +
 
 +
==== Seminar 8 (Bakhteev, [https://github.com/bahleg/tex_slides/raw/master/oct_19/slides2_hyper.pdf slides]) ====
 +
* Hyperparameter optimization
 +
* Bi-level optimization
 +
* RMD
 +
* Gradient optimization
 +
 
 +
==== Seminar 9 (Grabovoy, [https://github.com/andriygav/EMprior/blob/master/Lecture/Grabovoy2019EMprior.pdf slides]) ====
 +
* Mixture of Models
 +
* Mixture of Experts
 +
* Priors on the local Models
 +
 
 +
==== Seminar 10 (Isachenko, [[Медиа:Isachenko2019DeepGenerativeModels7.pdf‎|slides]]) ====
 +
* Reversible Residual Networks
 +
* Glow
 +
* Neural ODE
 +
 
 +
==== Seminar 11 (Bakhteev, [https://github.com/bahleg/tex_slides/raw/master/oct_19/slides3_meta.pdf slides]) ====
 +
* Meta-optimization
 +
* Pruning
 +
* Structure sampling
 +
 
 +
== Группа ==
 +
 
 +
=== 5 курс ===
 +
{|class="wikitable"
 +
|-
 +
! Студент
 +
! Тест 1
 +
! Тест 2
 +
! Тест 3
 +
! Тест 4
 +
! Тест 5
 +
! HW 1
 +
! HW 2
 +
|-
 +
|Васильев Илья
 +
| -
 +
| -
 +
| 0.59
 +
| -
 +
| -
 +
| -
 +
|
 +
|-
 +
|Гадаев Тамаз Тазикоевич
 +
| 0.56
 +
| 0.94
 +
| 0.75
 +
| -
 +
| 0.88
 +
| -
 +
|
 +
|-
 +
|Гладин Егор Леонидович
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
|
 +
|-
 +
|[[Участник:andriygav|Грабовой Андрей Валериевич]]
 +
| 0.63
 +
| 0.31
 +
| 0.67
 +
| 0
 +
| -
 +
|[[Media:Grabovoy2019Course5HM1.pdf‎|Essay]]
 +
|
 +
|-
 +
| Кислинский Вадим Геннадьевич
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
|
 +
|-
 +
| Козлинский Евгений Михайлович
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
|
 +
|-
 +
| Криницкий Константин Денисович
 +
| -
 +
| 0.25
 +
| -
 +
| -
 +
| -
 +
|[https://drive.google.com/file/d/1xk2uvz6FmWz0s1SCleF56pRFsTlXvQ8_/view?usp=sharing essay]
 +
|
 +
|-
 +
| Кириллов Егор Дмитриевич
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
|
 +
|-
 +
| Рогозина Анна Андреевна
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
|
 +
|-
 +
| Плетнев Никита Вячеславович
 +
| 0.82
 +
| 0.25
 +
| 0.67
 +
| -
 +
| 0.63
 +
|[https://drive.google.com/a/phystech.edu/file/d/1kkqHZDR2ATzWtf8FRc8BjbcNAXaApvy1/view?usp=drivesdk Essay]
 +
|
 +
|-
 +
| Малиновский Григорий Станиславович
 +
| 0.82
 +
| 0.81
 +
| 0.84
 +
| 1
 +
| 0.63
 +
||[https://drive.google.com/file/d/1UkmIDJgZEDTtzSk_a20IZlTy62F2tnJB/view?usp=sharing|task_1]
 +
|
 +
|-
 +
| Самохина Алина Максимовна
 +
| -
 +
| -
 +
| 0.25
 +
| 1
 +
| 0.75
 +
| -
 +
|
 +
|-
 +
| Султанов Азат Русланович
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
|
 +
|-
 +
| Федосов Павел Андреевич
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
| -
 +
|
 +
|-
 +
| Шульгин Егор Владимирович
 +
| -
 +
| -
 +
| 0.34
 +
| -
 +
| -
 +
| 0.13
 +
|
 +
|-
 +
|}
 +
 
 +
=== 6 курс ===
 +
{|class="wikitable"
 +
|-
 +
! Студент
 +
! HW 1
 +
! HW 2
 +
|-
 +
| Сайранов Данил
 +
| -
 +
|
 +
|-
 +
| Александра Гальцева
 +
| -
 +
|
 +
|-
 +
| Фельдман Даниил
 +
| -
 +
|
 +
|-
 +
| Никитин Филипп
 +
| -
 +
|
 +
|-
 +
| Фалахов И
 +
| -
 +
|
 +
|-
 +
| Собраков
 +
| -
 +
|
 +
|-
 +
|}
 +
 
 +
 
 +
* Topics
** Generative adversarial networks
** Generative adversarial networks
-
* Seminar 5 (Bakhteev)
 
** Methods of model selection
** Methods of model selection
** Generalization theorem
** Generalization theorem
-
* Seminar 6 (Bakhteev)
 
** Complexity theorems
** Complexity theorems
-
* Seminar 7 (Grabovoy?)
 
** Mixture of experts
** Mixture of experts
** Priors on the mixture
** Priors on the mixture
** Privileged learning and distilling
** Privileged learning and distilling
-
* Seminar 8 (Aduenko?)
 
** Theorem of number of experts
** Theorem of number of experts
-
* Seminar 9 (Vladimirova?)
 
** Prior propagation for deep learning networks
** Prior propagation for deep learning networks
-
* Seminar 10
 
** Directional Bayesian statistics
** Directional Bayesian statistics
-
* Seminar 11
 
** Bayesian structure learning
** Bayesian structure learning
-
* Seminar 12
 
** Probabilistic metric space construction
** Probabilistic metric space construction
-
* Seminar 13
 
** Informative prior
** Informative prior
-
* Seminar 14
 
** Bayesian programming
** Bayesian programming
-
 
+
** Informative prior with applications
-
 
+
-
 
+
-
 
+
-
 
+
-
 
+
-
 
+
-
 
+
-
 
+
-
 
+
-
* Informative prior with applications
+

Версия 20:02, 14 ноября 2019


Short link bit.ly/IS_B2

This series of seminars continues the course Bayesian model selection and investigates the theoretical aspects of model selection in various application problems.

Videolectures are available here.

Seminar 1 (Isachenko, slides)

  • Generative models
  • Applications
  • Autoregressive models (CharRNN, MADE, WaveNet, PixelCNN)

Seminar 2 (Isachenko, slides)

  • Generative vs discriminative
  • Latent variable models
  • Variational Inference
  • ELBO
  • Variational Autoencoder

Seminar 3 (Isachenko, slides)

  • Mean field approximation
  • Flow models (NICE, RealNVP)

Seminar 4 (Isachenko, slides)

  • VAE Limitations
  • Flows in VAE
  • Autoregressive flows (MAF, IAF, Parallel WaveNet)

Seminar 5 (Isachenko, slides)

  • IWAE (lower bound, posterior, inactive units)
  • ELBO surgery
  • VampPrior

Seminar 6 (Isachenko, slides)

  • Autoregressive decoder in VAE
  • Posterior collapse, decoder weakening
  • Disentangled representations
  • beta-VAE

Seminar 7 (Bakhteev, slides)

  • Model selection statement
  • ELBO for model selection
  • Early Stopping is Nonparametric Variational Inference
  • Langevin dynamics

Seminar 8 (Bakhteev, slides)

  • Hyperparameter optimization
  • Bi-level optimization
  • RMD
  • Gradient optimization

Seminar 9 (Grabovoy, slides)

  • Mixture of Models
  • Mixture of Experts
  • Priors on the local Models

Seminar 10 (Isachenko, slides)

  • Reversible Residual Networks
  • Glow
  • Neural ODE

Seminar 11 (Bakhteev, slides)

  • Meta-optimization
  • Pruning
  • Structure sampling

Группа

5 курс

Студент Тест 1 Тест 2 Тест 3 Тест 4 Тест 5 HW 1 HW 2
Васильев Илья - - 0.59 - - -
Гадаев Тамаз Тазикоевич 0.56 0.94 0.75 - 0.88 -
Гладин Егор Леонидович - - - - - -
Грабовой Андрей Валериевич 0.63 0.31 0.67 0 - Essay
Кислинский Вадим Геннадьевич - - - - - -
Козлинский Евгений Михайлович - - - - - -
Криницкий Константин Денисович - 0.25 - - - essay
Кириллов Егор Дмитриевич - - - - - -
Рогозина Анна Андреевна - - - - - -
Плетнев Никита Вячеславович 0.82 0.25 0.67 - 0.63 Essay
Малиновский Григорий Станиславович 0.82 0.81 0.84 1 0.63 [1]
Самохина Алина Максимовна - - 0.25 1 0.75 -
Султанов Азат Русланович - - - - - -
Федосов Павел Андреевич - - - - - -
Шульгин Егор Владимирович - - 0.34 - - 0.13

6 курс

Студент HW 1 HW 2
Сайранов Данил -
Александра Гальцева -
Фельдман Даниил -
Никитин Филипп -
Фалахов И -
Собраков -


  • Topics
    • Generative adversarial networks
    • Methods of model selection
    • Generalization theorem
    • Complexity theorems
    • Mixture of experts
    • Priors on the mixture
    • Privileged learning and distilling
    • Theorem of number of experts
    • Prior propagation for deep learning networks
    • Directional Bayesian statistics
    • Bayesian structure learning
    • Probabilistic metric space construction
    • Informative prior
    • Bayesian programming
    • Informative prior with applications
Личные инструменты