Winter Semester 2017/18

Seminar: Deep Learning (english)

(B_ECom, B_Inf, B_ITE, B_Tinf, B_WInf, B_Minf, B_STec, M_ECom, M_Inf, M_ITS, M_ITE)

Prof. Dr. Ulrich Hoffmann

November 17th (CHANGED: NO MORE ON November 18th) - HS5

The seminar will take place on Friday, November 17th (Excursion Day) (CHANGED: NO MORE ON the following Saturday, November 18th 2017) in Lecture Hall HS5.


November 17th, 2017 HS5

  • (canceled) - "Deep Feedforward Networks", Waldemar Husser
  • 09:30 - "Regularization for Deep Learning", Nicole Bruhn
  • (canceled) - "Optimization for Training Deep Models", Lea Morschel
  • 10:45 - "Convolutional Networks I", Sukanta Gosh
  • 12:00 - lunch break
  • 12:45 - "Recurrent and Recursive Nets II", Lukas Janssen
  • 14:00 - "Autoencoders", Marvin Saß


  • (canceled) - "Convolutional Networks II", Daniel Odesser
  • (canceled) - "Recurrent and Recursive Nets I", Varun Nayak
  • (moved - see above) - "Recurrent and Recursive Nets II", Lukas Janssen
  • (moved - see above) - "Autoencoders", Marvin Saß

Guests are welcome (Please register by email to

Zusätzliche Vorträge / Additional Talks

On Friday morning there will be another german talk prior to our seminar on mathematical topics non related to deep learning which are part of Prof. Iwanowski's seminar:

  1. "Das n-Damen-Problem auf dem verallgemeinerten Schachbrett", Florian Beenen

Participants are free visit these interesting talks as well.

Fixing the Schedule

The preparation meeting to fix the schedule of the talks took place on

Tuesday, October 17th 2017, 12:30 in seminar room SR9.

Topic Assignement

The preparation meeting introducing the different topics and assigning topics to participants took place on

     Monday, July 3rd 2017 at 12:30 p.m., seminar room SR9

Participation at this meeting is obligatory to attend the seminar. It is expected that you scan and skim the chapters of your interest in The Deep Learning book (see below).


Based on the Deep Learing Book [1][2] by Goodfellow, Benjio, and Courville we will work on different aspects of deep learning. During preparation all participants will famliarize with the basics of deep learning and then focus on a selected deep learning topic. These topics are presented in one hour talks. The following topics are available:

  1. Deep Feedforward Networks
  2. Regularization for Deep Learning
  3. Optimization for Training Deep Models
  4. Convolutional Networks
  5. Recurrent and Recursive Nets
  6. Linear Factor Models
  7. Autoencoders



Presenting each topic involves

  1. A talk (incl. slide presentation) of about 60 minutes plus 15 minutes time for discussion and methodological feed back. Talks are to be given in english.
  2. A report of approx. 20 pages that summarizes the topic in written form

Participation of attendees at every talk is required. Reasons for not showing up can exclusively be FH internal course conflicts or medical conditions. They still demand compensation. Not participating in two or more talks will result in failing the seminar.

Guests are welcome (Please register by email to