880008 :Deep Learning (CSAI/HAIT)

Algemeen

Voertaal Engels
Werkvorm: Lectures (Collegerooster)
Tentamenvorm: Presentation 20% and Paper 80%. There will be interim tests with which students can earn bonus points (to a max. of 20%) (Geen informatie over tentamendata bekend)
Niveau:Master
Studielast:6 ECTS credits
Inschrijving:Enrollment via Blackboard before start of lectures
Blackboard informatieLink to Blackboard (Als u de melding 'Guest are not allowed in this course' krijgt, dient u nog bij Blackboard in te loggen)

Docent(en)


prof. dr. E.O. Postma (coördinator)


Doel van de cursus

The objective of the course is to provide students with: (1) theoretical knowledge of convolutional and recurrent neural networks, (2) practical skills for performing experiments with deep learning, (3) understanding ways to (pre-)process signals, images, and texts for data science applications in general and for deep learning specifically.

 


Inhoud van de cursus

The lectures of the course start with a historical overview of deep learning. After a review of the formal basics, deep feedforward networks are explained in terms of (nonlinear) transformations and the backpropagation training procedure. Subsequently, the importance of regularization to prevent overfitting is explained and procedures for optimizing the induced model are outlined. Then, the notions of convolution and convolutional neural networks are explained. Sequence learning is addressed next by means of recurrent neural networks and their applications. This is followed by a review of the practical methodology of deep learning methods. Finally, applications are reviewed and recent scientific progress on the development of deep learning methods is discussed.
During skill classes, students are trained on applying the concepts addressed in the lectures.


Bijzonderheden

Deep Learning revolutionized machine learning by yielding the best performances in a large variety of application domains such as: speech recognition, image recognition, object detection, drug discovery and genomics. This course provides students with the understanding and skills to apply deep learning to signals, images, videos and textual sources. The course includes a training to run deep learning algorithms on special hardware.


Verplichte literatuur

  1. Goodfellow, I., Bengio, Y. and Courville, A., Deep Learning, Cambridge MA: MIT Press, 2016.


Gewenste voorkennis

880083 Machine Learning and 880022 Data Mining for Business and Governance


Vereiste voorkennis

880254 RS: Data Processing


Mogelijk interessant voor

  • Bedrijfscommunicatie en Digitale Media ( 2017 )
  • Communicatie-Design ( 2017 )
  • Communication and Information Sciences ( 2017 )
  • Cognitive Science and Artificial Intelligence ( 2017 )
  • Data Science: Business and Governance (voorjaar) ( 2017 )

(20-apr-2018)