|
Seminar Deep Learning Theory & Applications (SemDL)5 ECTS (englische Bezeichnung: Seminar Deep Learning Theory & Applications)
(Prüfungsordnungsmodul: Seminar im Masterstudium)
Modulverantwortliche/r: Andreas Maier Lehrende:
Andreas Maier, Stefan Evert, Vincent Christlein
Start semester: |
WS 2020/2021 | Duration: |
1 semester | Cycle: |
jährlich (WS) |
Präsenzzeit: |
30 Std. | Eigenstudium: |
120 Std. | Language: |
Englisch |
Lectures:
Inhalt:
Deep Neural Networks or so-called deep learning has attracted significant attention in the recent years. They have had a transformative influence on Natural Language Processing (NLP) and Artificial Intelligence (AI), with numerous success stories recent claims of superhuman learning performance in certain tasks. According to Young et al. (2017), more than 70% of the papers presented at recent NLP conferences made use of deep learning techniques.
Interestingly the concept of Neural Networks inspired researchers already over generations since Minky's famous book (cf. http://en.wikipedia.org/wiki/Society_of_Mind ). Yet again, this technology brings researchers to the believe that Neural Networks will eventually be able to learn everything (cf. http://www.ted.com/talks/jeremy_howard_the_wonderful_and_terrifying_implications_of_computers_that_can_learn ). This year's main topic is: "Multi-Task Learning for Document Analysis",
i.e. we will analyze Documents of different nature (text, images, etc.)
by means of multi-task learning using different techniques, such as
Natural Language Processing, Handwriting recognition, etc.
Lernziele und Kompetenzen:
The students
perform literature research based on a given scientific article
independently study the topic based on the found literature
present the given topic in a manner that is understandable for their peers
get to know the requirements of a scientific talk at an international conference
give a talk in English and gain language competence
investigate state-of-the-art deep neural network libraries
Literatur:
- Representation Learning: A Review and New Perspectives, Yoshua Bengio, Aaron Courville, Pascal Vincent, Arxiv, 2012.
Deep Learning, Yoshua Bengio, Ian Goodfellow, Aaron Courville, MIT Press
Goldberg, Yoav (2017). Neural Network Methods for Natural Language Processing. Number 37 in Synthesis Lectures on Human Language Technologies. Morgan & Claypool.
Young, Tom; Hazarika, Devamanyu; Poria, Soujanya; Cambria, Erik (2017). Recent trends in deep learning based natural language processing. CoRR, abs/1708.02709. http://arxiv.org/abs/1708.02709
Gradient-Based Learning Applied to Document Recognition, Yann Lecun, 1998
Dropout: A Simple Way to Prevent Neural Networks from Overfitting, Srivastava et al. 2014
Greedy layer-wise training of deep networks, Bengio, Yoshua, et al. Advances in neural information processing systems 19 (2007): 153.
Reducing the dimensionality of data with neural networks, Hinton et al. Science 313.5786 (2006): 504-507.
Training Deep and Recurrent Neural Networks with Hessian-Free Optimization, James Martens and Ilya Sutskever, Neural Networks: Tricks of the Trade, 2012.
Deep Boltzmann machines, Hinton et al.
Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion, Pascal Vincent et al.
A fast learning algorithm for deep belief nets, Hinton et al., 2006
ImageNet Classification with Deep Convolutional Neural Networks, Alex Krizhevsky, Ilya Sutskever, Geoffrey E Hinton, NIPS 2012
Regularization of Neural Networks using DropConnect, Wan et al., ICML
OverFeat: Integrated recognition, localization and detection using convolutional networks. Computing Research Repository, abs/1312.6229, 2013.
http://deeplearning.stanford.edu/wiki/index.php/UFLDL_Tutorial
http://deeplearning.net/tutorial/
Deep Learning Course on Coursera by Hinton
DL platform with GPU support: caffe, lasagne, torch etc.
Stanford University CS 224: Deep Learning for NLP (http://cs224d.stanford.edu )
University of Oxford: Deep Natural Language Processing (https://github.com/oxford-cs-deepnlp-2017/lectures )
Weitere Informationen:
Keywords: deep learning; neural networks; machine learning; pattern recognition; natural language processing
www: https://www5.cs.fau.de/lectures/ws-1920/seminar-deep-learning-theory-applications-semdl/
Studien-/Prüfungsleistungen:
Seminar Deep Learning (Prüfungsnummer: 514944)
(englischer Titel: Seminar Deep Learning)
- Prüfungsleistung, Seminarleistung, Dauer (in Minuten): 30, benotet, 5 ECTS
- Anteil an der Berechnung der Modulnote: 100.0 %
- weitere Erläuterungen:
Die Gesamtnote setzt sich zu 50% aus der Bewertung des Vortrags und zu 50% aus der Bewertung der Ausarbeitung / Implementierung zusammen. Ziel des Seminars ist die verständliche Aufbereitung eines Themas für andere Studierende. Die Vortragsdauer beträgt 30 Minuten. Ziel ist es, diese möglichst genau einzuhalten. Die Ausarbeitung umfasst 6 Seiten im Stil von IEEE-Konferenzbeiträgen. Vortrag und Ausarbeitung sollten auf Englisch erfolgen. Alternativ kann eine Demonstration implementiert werden. In diesem Fall umfasst die Ausarbeitung lediglich 3 Seiten.
- Prüfungssprache: Englisch
- Erstablegung: WS 2020/20211. Wdh.: keine Wiederholung, 2. Wdh.: keine Wiederholung
|
|
|