Intro to DL Blended: различия между версиями
Перейти к навигации
Перейти к поиску
Zimovnov (обсуждение | вклад) Нет описания правки |
Zimovnov (обсуждение | вклад) Нет описания правки |
||
| Строка 21: | Строка 21: | ||
In writing, theoretical questions, for instance: | In writing, theoretical questions, for instance: | ||
# SGD variations: Moment, RMSProp, Adam with explanation | # SGD variations: Moment, RMSProp, Adam with explanation | ||
# Description of backprop and proof of its efficiency | # Description of backprop and proof of its efficiency (линейное время работы) | ||
# Gradient of a dense layer in matrix notation (with proof) | # Gradient of a dense layer in matrix notation (with proof) | ||
# Typical CNN architecture, purpose of each layer, how to do backprop | # Typical CNN architecture, purpose of each layer, how to do backprop | ||
Версия от 12:07, 10 марта 2019
Course program:
https://www.hse.ru/data/2018/06/05/1150113338/program-2129241367-JndYcQjSAq.pdf
Grading:
Cumulative grade = 80% online course + 20% additional project
Final grade = 75% cumulative grade + 25% final exam
Additional project:
Homework with Kaggle competition: https://docs.google.com/document/d/1kTMYq21UFqZOqftjKAPq8G7RRkO7kX3MomsVIVhW830/edit?usp=sharing
Release date: 10-03-2019 16:00
Deadline: 24-03-2019 03:00
Exam:
In writing, theoretical questions, for instance:
- SGD variations: Moment, RMSProp, Adam with explanation
- Description of backprop and proof of its efficiency (линейное время работы)
- Gradient of a dense layer in matrix notation (with proof)
- Typical CNN architecture, purpose of each layer, how to do backprop
- Inception V3 architecture choices
- Description of auto-encoder, application to images
- Gradient of RNN cell (with proof)
Семинары:
1. Keras Tutorial https://colab.research.google.com/drive/1HoEsK580KAzMGuvFyYwUFdnRzuZ_hC13