# Deep Learning and Artificial Intelligence (WS 2018/19)

## News

- [24.04.2019] Examination
**inspection**for the**additional exam**takes place at**May 2nd**from**10 a.m. s.t. to 11 a.m.**in 157 (Oe. 67) - [25.02.2019] Examination inspection for the final exam takes place at March 6th from 2 p.m. s.t. to 3 p.m. in 157 (Oe. 67)
- [22.02.2019] Small updates/correction in the solutions for exercise 9-1 c) and 13-1 a).
- [19.02.2019] Based on your questions and feedback, we made some corrections to the exercise sheets/solutions. In particular:
- 3-2 a) correction of value p2
- 5-1 a) sigma' instead of sigma in the Jacobian matrix; expanded notation of result
- 7-2 b) X(UW)^T instead of XUW
- 7-2 d) np.transpose instead of np.reshape in pca function
- 11-2 a) clearer formulation of the backup strategies to be marked

- [11.02.2019] The date for the
**additional exam**is**April 16 at 2 p.m.**Details see below. - [06.02.2019] Please note the list of allowed aids and excluded topics for the final exam (see below, section Final Exam)
- [01.02.2019] No tutorial will take place next Monday (4th of February 2019).
- [01.02.2019] The solution for exercise 13 was updated. Moreover, we added a tar.gz file that should be unpackable without permission issues for Mac users.
- [25.01.2019] Exercise 13 was updated.
- [11.01.2019] The description of exercise 10-2a and the solution for 10-2c were updated.
- [07.12.2018] Exercise 07 (including solution) was updated due to notation clash.
- [04.12.2018] The date for the
**final exam**is**Feb 21 at 10 a.m.**Details see below. - [23.11.2018] Note that we made a small update regarding the notation in exercise 2 in exercise sheet 6. The latest version is now online.
- [20.11.2018] To further support your learning process, we decided to provide the solutions to the tutorial exercises from now on. They will be published after the corresponding tutorial lesson and are password-protected. Everyone registered to the course should have received an Email with the credentials to access the solutions. Please let the assistants of the lecture know if you are taking the course but did not receive an Email.
- [09.11.2018] Since there was some confusion about the indices in the cross-correlation/convolution formulas, we provide some additional explanations in the file `crosscors.pdf' (see solutions of tutorial 4). In the updated lecture slides, the formula for full padding is used.
- [09.11.2018] An updated version of the script on Convolutional Neural Networks with a correction of the convolution formula was uploaded. Please make sure you have the correct version.
- [06.11.2018] The solution to the coding exercise of sheet 3 is now online.
- [17.10.2018]
**Due to the amount of students actually visiting classes (about 25% of the registered master students), there is currently enough space in the classroom. Thus, we can reallow bachelor students to the course.**

Please, still keep in mind that this is still an advanced course which is not part of the bachelor curriculum and that prior knowledge in machine learning is still mandatory to understand the contents. - [16.10.2018]
~~Given the current amount of registered students, we cannot allow bachelor students to participate.~~ - [26.09.2018] Please note that
**"Machine Learning" or "Knowledge Discovery in Databases I" are essential**in order to follow this lecture. If you have not yet visited these lectures, please listen those first and visit "Deep Learning and Artificial Intellience" next year. Thank you for your understanding. - [22.08.2018] Registration is open via Uniworx.

## Organisation

**Umfang:**3+2 hours weekly (equals 6 ECTS)**Responsible Professor:**Prof. Dr. Matthias Schubert

**Lecturers:**Dr. Florian Büttner, Dr. Markus Geipel, Pankaj Gupta, Dr. Denis Krompass, Prof. Dr. Matthias Schubert, Dr. Sigurd Spieckermann, Prof. Dr. Volker Tresp

**Assistants:**Sebastian Schmoll, Sabrina Friedl**Required:**Lecture "Machine Learning" or equivalent, Lecture "Knowledge Discovery in Databases I" or equivalent**Audience:**The lecture is directed towards Master students in Mediainformatics, Bioinformatics, Informatics, and Data Science

## Final Exam

When: Thu, Feb 21 from 10 a.m. s.t. to 12 p.m.

Where: Main Building M 218

Registration: Uniworx

Allowed aids: One sheet of Din A4 paper with **handwritten** notes on both sides, calculator

List of topics not relevant for the exam: excluded_topics

## Additional Exam

When: Tue, April 16 from 2 p.m. s.t. to 4 p.m.

Where: Main Building M 018

Registration: Uniworx

Allowed aids: One sheet of Din A4 paper with handwritten notes on both sides, calculator

List of topics not relevant for the exam: excluded_topics

## Time and Locations

All times are c.t. (cum tempore)

Component |
When |
Where |
Starts at |

Lecture | Wed, 13.00 - 16.00 h | Room M 010 (HGB) | 17.10.2018 |

Tutorial 1 | Mo, 14.00 - 16.00 h | Room E 006 (HGB) | 22.10.2018 |

Tutorial 2 | Mo 16.00 - 18.00 h | Room E 006 (HGB) | 22.10.2018 |

## Content

During the last decade the availability of large amounts of data and the strong increase in computing power allowed a renaissance of neural networks and advanced planning techniques for independent agents. Whereas the area of deep learning extended well established neural network technology to allow a whole new level of data transformation, modern reinforcement learning techniques yield the artificial backbone for intelligent assistant systems and autonomous vehicles.

The course starts with an introduction to neural networks and explains the developments that led to deep architectures. Furthermore, the course gives an introduction to advanced planning techniques and how they can be trained using deep neural networks and other machine learning technologies.

## Course Schedule

Lecture | Tutorial | |||
---|---|---|---|---|

Date | Topic | Date | Exercises | Solutions |

17.10.2018 | IntroductionSpeaker: Sigurd Spieckermann |
22.10.2018 | Python Introduction (ex01.html, ex01.ipynb, ex01.py) | |

24.10.2018 | Basic Neural Networks Speaker: Florian Büttner |
29.10.2018 | Math Primer (ex02) | ex02_sol |

31.10.2018 | Training Neural NetworksSpeaker: Sigurd Spieckermann |
05.11.2018 | Computational Graphs and Vanishing Gradients (ex03, ex03.ipynb) | ex03_sol, ex03_sol.ipynb |

07.11.2018 | Convolutional Neural NetworksSpeaker: Denis Krompass |
12.11.2018 | Convolutional Neural Networks (ex04, ex04.ipynb) | ex04_sol, crosscors |

14.11.2018 | Recurrent Neural NetworksSpeaker: Pankaj Gupta |
19.11.2018 | RNNs (ex05), CIFAR10 Competition (CIFAR10.ipynb) |
ex05_sol |

21.11.2018 | Deep Learning and UncertaintySpeaker: Markus Geipel |
26.11.2018 | Uncertainty (ex06, MCMC.ipynb, LSTM.ipynb) | ex06_sol, MCMC_sol.ipynb, LSTM_sol.ipynb |

28.11.2018 | Representation and Distributional LearningSpeaker: Pankaj Gupta |
03.12.2018 | Representation & Distributional Learning (ex07, autoencoder.ipynb, RBM.ipynb) | ex07_sol, autoencoder_sol.ipynb |

05.12.2018 | Deep Learning ToolsSpeaker: Denis Krompass |
10.12.2018 | Tooling (ex08, torch.ipynb) | ex08_sol, Autoencoder.py, AutoencoderKeras.py |

12.12.2018 | Generative ModelsSpeaker: Florian Büttner |
17.12.2018 | Generative Models (ex09, vae.ipynb) | ex09_sol |

19.12.2018 | Sequential Decision Problems and autonomous AgentsSpeaker: Matthias Schubert |
07.01.2018 | MDP (ex10, mdp.ipynb) | ex10_sol mdp_sol.ipynb |

09.01.2019 | Model-free Reinforcement LearningSpeaker: Matthias Schubert |
14.01.2019 | RL (ex11, rl.ipynb) | ex11_sol rl_sol.ipynb |

16.01.2019 | Value Function ApproximationSpeaker: Matthias Schubert |
21.01.2019 | Function Approximation (ex12, QLearning.ipynb) | ex12_sol QLearning_sol.ipynb |

23.01.2019 | Policy Gradients and Actor Critic LearningSpeaker: Matthias Schubert |
28.01.2019 | Policy Gradients (ex13, Reinforce.zip Reinforce.tar.gz) |
ex13_sol Reinforce_sol.zip Reinforce_sol.tar.gz |

30.01.2019 | Knowledge Graphs in AI Speaker: Volker Tresp |
04.02.2019 | No tutorial (Pong competition, pong_v2) |
no one submitted... |

06.02.2019 | Q&A | ---------------- |