Download Deep Learning - studiegids UGent

Course
Specifications
Valid as from the academic year 2016-2017
Deep Learning (F000884)
Course size
Credits 4.0
(nominal values; actual values may depend on programme)
Study time 120 h
Contact hrs
40.0 h
Course offerings and teaching methods in academic year 2017-2018
A (semester 2)
seminar: practical PC room classes
20.0 h
demonstration
2.5 h
group work
5.0 h
lecture
10.0 h
project
2.5 h
Lecturers in academic year 2017-2018
Shanahan, James
EB07
lecturer-in-charge
Offered in the following programmes in 2017-2018
crdts
Master of Science in Business Engineering (main subject Data
4
Analytics)
Master of Science in Business Engineering (main subject Finance)
4
Master of Science in Business Engineering (main subject Operations 4
Management)
Master of Science in Marketing Analysis
4
offering
A
A
A
A
Teaching languages
English
Keywords
Deep learning, artificial neural networks, TensorFlow, artificial intelligence
Position of the course
We want to offer courses to "Master of Science in Marketing Analysis" students that
reflect the state-of-the-art in research methodology. Deep Learning is an example of
such an emerging domain. Bill Gates: "AI is the holy grail".
Contents
Deep learning represents an entirely new approach to artificial intelligence. It tries to
learn from experience and understand the world in terms of a hierarchy of concepts,
with each concept defined in terms of its relation to simpler concepts. This approach
avoids having to formally specify all of the knowledge that the system needs. Deep
learning builds on artificial neural networks, hence, this course starts out with an indepth explanation of ANNs. This course details 1. The benefits of neural networks over
other learning algorithms; 2. The benefits of “deep” neural networks over “shallow”
architectures.
The course will detail marketing cases using deep learning in the field of DL for NLP
(natural language processing) with application to sentiment analysis and DL for image
processing (e.g., using pictorial stimuli in predictive marketing models).
Case studies will make use of TensorFlow.
Initial competences
CRISP-DM data mining methodology.
Programming skills.
Final competences
1 Determining when and how to use Deep Learning for solving complex marketing
(Approved)
1
1 problems.
2 Using and levering complex data (e.g., pictures, audio, video).
3 Solving business problems using Deep Learning.
4 Validating the results of one's own research with existing literature.
Conditions for credit contract
This course unit cannot be taken via a credit contract
Conditions for exam contract
This course unit cannot be taken via an exam contract
Teaching methods
Demonstration, group work, lecture, project, seminar: practical PC room classes
Learning materials and price
Own syllabus
Scientific papers:
• Bengio, Yoshua; LeCun, Yann; Hinton, Geoffrey (2015). "Deep
• Learning". Nature 521: 436–444.
• Vilnai-Yavetz I., Tifferet S. (2015), "A Picture Is Worth a Thousand Words:
• Segmenting Consumers by Facebook Profile Images", Journal of Interactive
• Marketing, 32: 53-69.
References
Goodfellow I, Bengio Y., Courville A. (2016), "Deep Learning", MIT Press.
Course content-related study coaching
Numerous exercises are being solved during sessions. In addition, assignments (to be
solved in teams) are handed out.
Students receive coaching in the process of solving the assignments and feedback
afterwards (collectively, by team and individually).
Evaluation methods
continuous assessment
Examination methods in case of periodic evaluation during the first examination period
Examination methods in case of periodic evaluation during the second examination period
Examination methods in case of permanent evaluation
Written examination with open questions, open book examination, oral examination,
assignment, peer assessment, report
Possibilities of retake in case of permanent evaluation
examination during the second examination period is possible in modified form
Calculation of the examination mark
Written part: 80%
Oral part: 20%
potentially adjusted by peer assessment. (Approved)
2