Course Syllabus

78342 (Section G), 78341 (U)

Neural Network

 

Content

We will understand a number of issues about Deep Neural Networks.  Our goal is to understand the mathematical underpinning of a number of tools in Deep Learning.  We hope that this will give you the foundation to understand and critically think about continued advances in this dynamic area.

List of Topics
Linear Regression
Logistic Regression
Elementary Logic
Backpropagation
Gradient Descent
Feedforward Networks
Testing, Validation, and Training
Feature Importance
Dropout
Batch Normalization
Convolutional Neural Networks
Recurrent Neural Networks
Reinforcement Learning
Generative Adversarial Networks

 

  • Instructor:  Richard Sowers
  • Class Meets:  08:00AM – 09:20AM TR in 1310 DCL.
  • Office hours: TBA in 201D Transportation (in person) and on Zoom (@illinois.edu login required) 
  • TA's:  See canvas announcement.

Assessments

 

Component Date/Frequency Percentage
Midterm 1 October 14 15%
Midterm 2 December 9 15%
Individual Theory Assignments Each module (roughly) 25%
Group Coding Assignments Each module (roughly) 25%
Deep Dive Project End of Semester 20%

 

  • Group Coding Assignments:
    • We will use Google Drive and Google Colab (Python Jupyter notebooks) for teaching and the Group Coding Assignments.  You can access these through Google Apps @ Illinois; you need to have Account Status “On” for Google Apps at https://cloud-dashboard.illinois.edu/cbdash/ and then log in via g.illinois.edu
    • Groups will be randomly set by canvas.  Groups will be re-randomized several times. 
    • Submissions will be via URL to your Google Colab file.  Make sure to set the permissions to "Google Apps @ Illinois". 
      • Best practices seems to be to set permissions to "Google Apps @ Illinois".  That means that the TA and I can automatically access your notebook.  The actual URL is long and random enough that others probably can't guess it.
      • Access for data files should be "Anyone with the link"; that ensures that colab notebooks can access it.  See lecture notebooks for examples.
    • The equivalent of one Group Coding Assignment will be for "longitudinal team effort".   You will get full credit for this unless there is a pattern of not contributing to groups.
  • Midterm Exams: These will be held in the regular classroom.
  • Extra Credit: You will get up to 2% extra credit for attendance.   Attendance will be taken randomly at any time during the class.
  • Late work:  5% per day will be deducted for late homework.  No exceptions.
  • Late Enrollment:  Email me if you join the class after the first 2 weeks.  I will excuse your assignments due on or before the second day after you join the class.

 

Strategy for Learning the Material

My (PDF) lecture notes (available in advance of each lecture) and verbal lectures are designed to satisfy slightly different constraints.  The lecture notes are constrained by space, and the desire to, as much as possible, have complete ideas on each slide.  Verbal lectures often explain motivations, connections, and interpretations.  Perhaps the best way to learn the material is to

  1. Quickly skim through the lecture notes in advance of the class.  By doing so, you get an idea of where the verbal lecture will be heading
  2. During the verbal lecture, annotate (either on a printed out copy or on a tablet) the lecture notes.  The goal is to jot down ideas, connections, and interpretations so that you can remember them after the lecture.
  3. Review the material after the lecture to solidify your understanding of both details and big-picture ideas. 

Miscellanea