Instructor:      Jeffrey Horn,  email:     jhorn@nmu.edu
                            office:  2202 JXJ
                            phone:  227-1607
Office Hours:    TBA

Textbook:   None required
Classroom:  John X. Jamrich (JXJ) 1317
Meeting Times:  4:00-4:50pm  Tu Th NMU CRN:  12168

Our web page:   http://euclid.nmu.edu/~jeffhorn/Classes/DeepLearning/Winter2017

 (I will use the web page EXTENSIVELY, posting everything I can up there as soon as possible.  This includes all electronic forms of handouts, assignments, solutions, sample tests, etc.  Also I will post announcements, links to interesting, topic-related sites, etc.  So please check our page regularly!  At the very least, once a week.   I will use your nmu email address for notices, and our class EduCat page for grade posting.)

COURSE OBJECTIVES:

Students will develop a practical understanding of deep learning (DL) neural networks, sufficient to indendently (1)  apply DL to real world problems, (2) explore further concepts in DL, and (3) maintain currency with new techniques, algorithms, and software tools for DL.

LEARNING OUTCOMES:

TOPICS:  (tentative)

OVERVIEW

THE PERCEPTRON

BACKPROPAGATION:  MULTI-LAYER Neural Networks (NNs)

CONVOLUTIONAL NNs

RECURRENT NNs

DEEP NNs

20%  Homeworks. 
20%  Quizzes.  
20%  Topic exams.
20%  Final exam, comprehensive.
20%  Final Project

Late Policy:  For homeworks, 5% off for each day late (counting only days that the university is open; e.g., not weekends or snow days).   But of course I cannot accept them after solutions are handed out!  As for exams, those cannot be made up except under the most severe and extenuating emergencies!  Don't take a chance if you don't have to! 

COMPUTING FACILITIES:

Deep learning is computationally intensive.   Thousands to millions of training instances are needed, which are often images.   The neural networks themselves often contain thousands of "neurons" with hundreds of thousands of connections.   For small models we can use our laptops (a.k.a., notebook PCs), but our goal is learn practical, real world concepts.   For this we will need to use dedicated high-performance computing (HPC) assets, such as (1) NMU's own cluster computer, the Williac, (2) GPU computation such as the nVidia GeForce 970 in the NERL, or (3) cloud-based HPC.

If you have a need for disability-related accommodations or services, please inform the Coordinator of Disability Services in the Dean of Students Office at 2101 C. B. Hedgcock Building (227-1700 or disserv@nmu.edu). Reasonable and effective accommodations and services will be provided to students if requests are made in a timely manner, with appropriate documentation, in accordance with federal, state, and University guidelines.

LAPTOP (and mobile device) CLASSROOM USE POLICY

As a computer scientist I am of course devoted to mobile computing and communication devices (that is, gadgets).  However, this course involves some deep concepts concerning the nature of computation, information, and communication, and indeed even human nature.  My experience is that most people, myself most especially, cannot be expected to really grasp these concepts in the classroom without long periods of unbroken attention.  As much as I value electronic-based multi-tasking (and I do think that we need to multi-task throughout the modern day), I have now come to the conclusion that certain insights into the universe cannot be conveyed while multi-tasking.  Period. 

Therefore, my policy is to allow laptop use during certain portions of the class period, as I announce them.  So there will be times, every class meeting, when I will demand that all laptop covers be closed (at least lowered so that they cannot be seen by anyone).   Such times will last perhaps 20-30 minutes, after which students will be allowed to open up and use their laptops for class exercises or lab work (the current assignment).  If a student absolutely needs to use his or her laptop for note-taking during the "deep lectures," then he or she can request individual permission from me. 

As for cell phones, PDAs, handheld game-consoles, iPods, etc., use of such devices will also not be allowed during "deep lecture" (cell phones can be set to "vibrate" or to some other inaudible notification mode) for receiving emergency calls.   There will be break times of five to ten minutes every half hour or so for students to check for messages, make short calls, etc.