Lecture: Tuesday, 16:15-17:45 in 32/110
Homework Tutorial: Wednesday, 12:15-13:45 in 35/E01
Open Homework Q&A Session: Friday, 14.15-15.45 in 35/E16
This course gives 8 ECTS. For that you have to pass 8 homeworks, pass the midterm exam and submit a final project at the end of the semester.
| Date | Topic | Lecture | Homework | Solutions | |:------:|:------------:|:-----------:|:----------:|:-------:|:-------:| | 29.01. - 05.11.| Introduction | session01.pdf | homework01.pdf | - | | 05.11. - 12.11.| Perceptron, MLP | session02.pdf | homework02.ipynb | homework02-solution.ipynb | | 12.11. - 19.11.| Backpropagation | session03.pdf | homework03.ipynb | homework03-solution.ipynb | | 19.11. - 26.11.| TensorFlow | session04.pdf session04.ipynb | homework04.ipynb | homework04-solution.ipynb | | 26.11. - 03.12.| More on Deep Neural Networks | session05.pdf | homework05.ipynb | homework05-solution.ipynb | | 03.12. - 10.12.| Convolutional Neural Networks | session06.pdf |homework06.ipynb | homework06-solution.ipynb | | 10.12. - 16.12.| Training DNNs | session07.pdf | homework07.ipynb | n.a. | | TBA | Midterm Exam | - | - | - | | 18.12. - 07.01.| Christmas Break | - | - | - | | 07.01. - 14.01.| Advanced CNNs + Visualization | session08.pdf | homework08.ipynb| homework08-solution.ipynb | | 14.01. - 21.01.| Recurrent Neural Networks | session09.pdf | homework09.ipynb | homework09-solution.ipynb | | 21.01. - 28.01.| Word Embeddings | session10.pdf | homework10.ipynb | homework10-solution.ipynb | | 28.01. - 04.02.| Generative Models| session11.pdf |homework11.ipynb | homework11-solution.ipynb | | 04.02. - 11.02.| Deep Reinforcement Learning | only on Stud.IP | homework12.ipynb | n.a. | | 11.02. - 14.02.| Practical Aspects |session13.pdf | n.a. | n.a. | | 14.02. | Deadline Fix Topic for Final Project | - | - | - | | 30.03. | Deadline Final Project | - | - | - |
The lecture is recorded weekly. These recordings are only available via Stud.IP.
The homework will be published on Tuesdays after the lecture. The submission deadline is Tuesdays 14.00.
You have to pass 8 out of 10-12 homeworks. There are four posssible ratings: outstanding, done, not done, fail. All ratings, but fail, are a pass!
When you submit your homework you have to rate your homework yourself. Each rating has a different meaning and a different consequence.
outstanding - Your homework runs perfectly and you have put effort into making the code beautiful and readable. Other students would benefit from reading your code. (more info)
We will check this. If we approve you get +0.05 on your final grade and we will publish the homework such that others can have a look at it.
done - Your homework runs and it does what it's supposed to do.
We will randomly check this.
not done - You did not solve the homework, but you really put in an effort.
In this case we expect you to include a statement explaining your problems and your tried solutions in your homework submission. Additionally you have one week for uploading a second document in which you explain whether the published solutions resolved your problems or what the remaining problems are. We will check this and if we approve that you put in enough effort you will get a pass.
fail - You did not submit anything or you did not finish the homework, but did not put in enough effort for the not done rating.
Homework submission - You can submit your homework via Colab or via Stud.IP. Colab submission instructions: colab-intro.pdf . Stud.IP submission instructions: Submit your homework to your group folder following the naming convention: G{group_number}_H{homework_number}_{rating}.ipynb
where rating is O, D, ND or F. (e.g. G01_H01_ND.ipynb).
Cheating - Please be honest. If we have to downgrade the rating, you will get an Email in which we will explain what was missing. However if you constantly or clearly give the wrong rating you will get a Fail. Further if you submit a homework as Done that is not running you will get a Fail and we will substract one Outstanding.
The midterm exam will cover all contents of the first seven sessions. You won't have to code although there might be simple questions about code in the exam. The exam should be easily doable if you follow the course (that is visiting the lectures and doing the homework).
In the final project you will choose a publication or a task for which you will implement an artificial neural network in TensorFlow. Additionally you will write a blog-post that explains your project. If you re-implement a publication you roughly explain the theoretical background. If you solve a task you explain how you went about solving it. In both cases you explain your implementation. Check the proposal list for possible topics. This list will be constantly updated. (more infos follow)
In case of general or organizational issues contact me: leffenberger@uos.de.
In case of content questions contact Sahar or Leon: sniknam@uos.de, lschmid@uos.de.
I would like to thank Lukas Braun, who when he was a Bachelor student at the University of Osnabrueck, designed the outlines of this course.