Chris Pollett> CS256
( Print View )

Student Corner:
[Submit Sec2]
[Grades Sec2]

[Lecture Notes]
[Discussion Board]

Course Info:
[Texts & Links]
[Description]
[Course Outcomes]
[Outcomes Matrix]
[Course Schedule]
[Grading]
[Requirements/HW/Quizzes]
[Class Protocols]
[Exam Info]
[Regrades]
[University Policies]
[Announcements]

HW Assignments:
[Hw1] [Hw2] [Hw3]
[Hw4] [Hw5] [Quizzes]

Practice Exams:
[Midterm] [Final]

CS256 Fall 2021 Sec2 Home Page/Syllabus

Topics in Artificial Intelligence

Instructor: Chris Pollett
Office: MH 214
Phone Number: (408) 924 5145
Email: chris@pollett.org
Office Hours: MW 2:45-3:45pm in MH 214
Class Meets:
Sec2 MW 4:00-5:15pm SCI 311 (in-person)

Prerequisites

To take this class you must have taken:
CS156
with a grade of C- or better.

Texts and Links

Required Texts: Deep Learning. Ian Goodfellow, Yoshua Bengio, and Aaron Courville. MIT Press. 2016.
Online References and Other Links: Python.
Keras.
Tensorflow.
Pytorch.

Description

Artificial Neural Networks provide a general model for computers to learn how to perform a variety of tasks. With the advent of both cheap hardware, and large scale data sets, it has become increasingly possible to train computers to perform tasks that previously could only be carried out by humans, if at all. In this class, we will take both a pragmatic and theoretical approach to how to design and build artificial neural networks. We will start with an overview of machine learning and the math needed for machine learning. We will discuss single neuron/layer networks (perceptron, SVM, etc) and how to train them. We will then look at multi-layer networks and variations on the back-propagation training algorithm. We will consider different kinds of neural networks layer types, such as convolutional neural network layer, recurrent neural network layers, etc and when and why we would use these different kinds of layers. To make our models generalize we will consider different a variety of regularization techniques. As the class goes on, we will consider how to design neural networks for different applications such as vision tasks like object classification and detection, language processing, planning tasks, and policy selection. We will also learn how to tune our networks and determine how data we need to be able to effectively train them. Code for this class will be in Python. For our more sophisticated networks we will use libraries such as Pytorch, Keras, Tensorflow, etc.

Course Learning Outcomes (CLOs)

By the end of this course, a student should be able to:

CLO1 -- Be able to code without a library a single perceptron training algorithm.

CLO2 -- Be able to predict the effect of different activation functions on the ability of a network to learn.

CLO3 -- Be able to explain how different neural network training algorithms work.

CLO4 -- Be able to select neural network layers type to build a network suitable for various learning tasks such as object classification, object detection, language processing, planning, policy selection, etc.

CLO5 -- Be able to select an appropriate regularization technique for a given learning task.

CLO6 -- Be able to code and train with a library such as Tensorflow or Pytorch a multi-layer neural network.

CLO7 -- Be able to measure the performance of a model, determine if more data in needed, as well as how to tune the model.

Course Schedule

Below is a tentative time table for when we'll do things this quarter:

Week 1:Aug 23, Aug 25 Read Ch 1 Intro, Skim Ch 2 to 5 (over next few weeks), Background Probability
Week 2:Aug 30, Sep 1 Background linear algebra, start Perceptrons
Week 3:Sep 6 (No Class), Sep 8 PAC Learning, Start Python
Week 4:Sep 13, Sep 15 Python, Ch 5
Week 5:Sep 20(Hw1), Sep 22 Perceptron Networks, SVMs
Week 6:Sep 27, Sep 29 Numpy, Pillow, Neural Net Experiments, Ch 6 Feedforward Networks
Week 7:Oct 4, Oct 6 More Ch 6 Cost Functions, Output Layers, Minimization Methods, Hidden Units, Stochastic Gradient Descent
Week 8:Oct 11(Hw2), Oct 13(Midterm) Review
Week 9:Oct 18, Oct 20 Finish Ch 6, Backpropagation. Tensorflow, Pytorch, Keras. Ch 7 Regularization
Week 10:Oct 25, Oct 27 Finish Ch 7, Ch 8 Learning Algorithms as an Optimization Problem
Week 11:Nov 1(Hw3), Nov 3 Finish Ch 8, Ch 9 Convolutional Networks
Week 12:Nov 8, Nov 10 Ch 10 Recurrent and Recursive Networks
Week 13:Nov 15, Nov 17 Ch 11 Tuning and Debugging Networks, Ch 12 Neural Net Applications
Week 14:Nov 22, Nov 24 (No Class) Ch 14 Autoencoders
Week 15:Nov 29, Dec 1 Recent advances in network topologies and training
Week 16:Dec 6(Hw5) Review
The final will be Wednesday, December 8 from 2:45-5:00 PM.

Grading

HWs and Quizzes 50%
Midterm 20%
Final 30%
Total100%

Grades will be calculated in the following manner: The person or persons with the highest aggregate score will receive an A+. A score of 55 will be the cut-off for a B-. The region between this high and low score will be divided into five equal-sized regions. From the top region to the low region, a score falling within a region receives the grade: A, A-, B+, B, B-. If the boundary between an A and an A- is 85, then the score 85 counts as an A-. Scores below 55 but above 50 receive the grade D. Those below 50 receive the grade F.

If you do better than an A- in this class and want me to write you a letter of recommendation, I will generally be willing provided you ask me within two years of taking my course. Be advised that I write better letters if I know you to some degree.

Course Requirements, Homework, Quiz Info, and In-class exercises

This semester we will have five homeworks, weekly quizzes, and weekly in-class exercises.

Every Monday this semester, except the first day of class, the Midterm Review Day, and holidays, there will be a quiz on the previous week's material. The answer to the quiz will either be multiple choice, true-false, or a simple numeric answer that does not require a calculator. Each quiz is worth a maximum of 1pt with no partial credit being given. Out of the total of twelve quizzes this semester, I will keep your ten best scores.

On Wednesday's, we will spend 15-20 minutes of class on an in-class exercise. You will be asked to post your solution to these exercises to the class discussion board. Doing so is worth 1 "insurance point/pre-point" towards your grade. An "insurance point/pre-point" can be used to get one missed point back on a midterm or final, up to half of that test's total score. For example, if you scored 0 on the midterm and have 10 insurance points, you can use your insurance points, so that your midterm score is a 10. On the other hand, if you score 18/20 on the midterm, you can use at most 1 insurance point since half of what you missed (2pts) on the midterm is 1pt.

Links to the current list of homeworks and quizzes can be found on the left hand side of the class homepage. After an assignment has been returned, a link to its solution (based on the best student solutions) will be placed off the assignment page. Material from assignments may appear on midterms and finals. For homeworks you are encouraged to work in groups of up to three people. Only one person out of this group needs to submit the homework assignment; however, the members of the group need to be clearly identified in all submitted files.

Homeworks for this class will be submitted and returned completely electronically. To submit an assignment click on the submit homework link for your section on the left hand side of the homepage and filling out the on-line form. Hardcopies or e-mail versions of your assignments will be rejected and not receive credit. Homeworks will always be due by midnight according to the departmental web server on the day their due. Late homeworks will not be accepted and missed quizzes cannot be made up; however, your lowest score amongst the five homeworks and your quiz total will be dropped.

When doing the programming part of an assignment please make sure to adhere to the specification given as closely as possible. Names of files should be as given, etc. Failure to follow the specification may result in your homework not being graded and you receiving a zero for your work.

Classroom Protocol

I will start lecturing close to the official start time for this class modulo getting tangled up in any audio/visual presentation tools I am using. Once I start lecturing, please refrain from talking to each other, answering your cell phone, etc. If something I am talking about is unclear to you, feel free to ask a question about it. Typically, on practice tests days, you will get to work in groups, and in so doing, turn your desks facing each other, etc. Please return your desks back to the way they were at the end of class. This class has an online class discussion board which can be used to post questions relating to the homework and tests. Please keep discussions on this board civil. This board will be moderated. Class and discussion board participation, although not a component of your grade, will be considered if you ask me to write you a letter of recommendation.

Exams

The midterm will be during class time on: Oct 13.

The final will be: Wednesday, December 8 from 2:45-5:00 PM.

All exams are closed book, closed notes and in this classroom. You will be allowed only the test and your pen or pencil on your desk during these exams. The final will cover material from the whole semester although there will be an emphasis on material after the last midterm. No make ups will be given. The final exam may be scaled to replace a midterm grade if it was missed under provably legitimate circumstances. These exams will test whether or not you have mastered the material both presented in class or assigned as homework during the quarter. My exams usually consist of a series of essay style questions. I try to avoid making tricky problems. The week before each exam I will give out a list of problems representative of the level of difficulty of problems the student will be expected to answer on the exam. Any disputes concerning grades on exams should be directed to me, Professor Pollett.

Regrades

If you believe an error was made in the grading of your program or exam, you may request in person a regrade from me, Professor Pollett, during my office hours. I do not accept e-mail requests for regrades. A request for a regrade must be made no more than a week after the homework or a midterm is returned. If you cannot find me before the end of the semester and you would like to request a regrade of your final, you may see me in person at the start of the immediately following semester.

University Policies and Procedures

Per University Policy S16-9, university-wide policy information relevant to all courses, such as student class time requirements expectations, academic integrity, accommodations, etc. will be available on Office of Graduate and Undergraduate Programs' Syllabus Information web page at http://www.sjsu.edu/gup/syllabusinfo/. Below are some brief comments on some of these policies as they pertain to this class.

Academic Integrity

For this class, you should obviously not cheat on tests. For homeworks, you should not discuss or share code or problem solutions between groups! At a minimum a 0 on the assignment or test will be given. A student caught using resources like Rent-a-coder will receive an F for the course. Faculty members are required to report all infractions to the Office of Student Conduct and Ethical Development.

Accommodations

If you need a classroom accommodation for this class, and have registered with the Accessible Education Center, please come see me earlier rather than later in the semester to give me a heads up on how to be of assistance.