CSE 151 is an upper-division course devoted to the basic concepts and algorithms of modern artificial intelligence (AI), especially machine learning. 151 is part of a two quarter sequence with CSE 150, but each course may be taken independently. 151 will focus on statistical reasoning and learning, while 150 covers search and logical reasoning methods. Both courses cover theory and applications.
Undergraduate and M.S. students in CSE, mathematics, engineering, and cognitive science are welcome in 151. The only prerequisite for 151 is CSE 100 (upper-division data structures) or equivalent experience. For registration, the section id is 631645. Note: If you cannot register for any reason, do not worry. Come to the first lectures and the instructor will sign an add card.
Some specific AI topics that will be covered in CSE 151 are:
The instructor is Charles
Elkan,
Professor. Lectures will be on Tuesdays and Thursdays from 9:30am to 10:50am in
room 2154 in the CSE building.
Lecture notes for each class meeting will be published on this website.
There will be a weekly one-hour section every Monday at 11am, led by the teaching assistant, who is Albert Park.
| September 25 |
Framework for AI, learning, and classifier learning. Nearest-neighbor classifiers. |
Assignment 1 |
| September 30 | Notion of optimal error rate. Theorem on error rate of 1NN. |
|
| October 2 | Linear classifiers, hyperplanes, perceptron algorithm. | |
| October 7 | Proof of convergence of perceptron algorithm. | |
| October 9 | Discrete probabilities, Bayes' rule, example of Bayesian medical diagnosis | Assignment 2 |
| October 14 | Using Bayes' rule for classification, naive Bayes (NB) simplifying assumption | |
| October 16 | Algorithm for NB training and prediction, spam example, time and space complexity | Assignment 3 |
| October 21 | Evaluation of classifiers, cross-validation, precision and recall | |
| October 23 | Unsupervised learning, principle of maximum likelihood, Bernoulli example | |
| October 28 | In-class midterm exam | |
| October 30 | Maximum likelihood for multinomial distributions | |
| November 4 | "Bag of words" representation for documents. Multinomial distribution to model sets of documents. | Assignment 4 |
| November 6 | Mixture models, mixture of multinomials, expectation-maximization (EM), initialization | |
| November 11 | No class because of Veterans' Day | |
| November 13 | EM review, deterministic annealing | Midterm solutions |
| November 18 | Markov decision processes, value iteration | Assignment 5 |
| November 20 | Policy iteration algorithm, Q-learning | Russell and Norvig notes |
| November 25 | Assignment 2 solutions |
| September 29 | Matlab basic commands and functions. Vectorization for efficiency. |
| October 6 | History of machine learning and artificial intelligence |
| October 13 | Review of perceptron convergence theorem. Conditional probabilities |
| October 20 | Project grading, Matlab performance tutorial |
| October 27 | Review session to prepare for the midterm |
| November 3 | Maximum likelihood example: the Poisson distribution |
| November 10 | No section |
| November 17 | Section will be rescheduled |
The course will have an open-book final examination on Thursday
December 11 from 8am to 11am. There will be one
midterm exam, also open-book, on Tuesday October 28 in class.
Examinations will be based
mainly on material from lectures, as summarized in the online lecture notes. Exam
questions may also involve knowledge from the assignments.
Open-book means that you may bring (i) your own personal handwritten notes, (ii) lecture notes from this website, (iii) Wikipedia pages and other documents distributed in class, or linked to from lecture notes, and (iv) a basic calculator. You may not bring books, copies of notes written by other people, or a computer.
There will be two or three written assignments, and three or four programming assignments. You may do each of the assignments either by yourself, or with exactly one partner, at your choice; you may change partners between assignments. The assignments are always due in class, i.e. at 9:30am on the due date. The penalty for late submission is 20% of the maximum score per day or part of day late. One day late means after 9:35am and before midnight on the due date, two days late means the next day, and so on. Generally, you will have one week for written assignments and two weeks for programming assignments.
Complete academic honesty is always required. For all programming assignments, code quality is very important. This includes useful commenting, clarity and concision, modularity and organization, and appropriate error checking. The first programming assignment is due on Thursday October 9.
The midterm will count for 10% of your overall grade, the final for 30%, each written assignment for 5%, and each programming assignment for 15%. (5% of your grade may be a gimme :-) There is no a priori correspondence between letter grades and numerical scores on the assignments or on the exams. You can evaluate your performance in the class by comparing your scores with the means and standard deviations, which will be announced. However there is also no fixed correspondence between letter grades and standard deviations above or below the mean. If all students do well in the absolute, then all students will get a good grade.
You should not drop CSE 151 just because you are unhappy with a score that you receive. Instead, you should make an appointment to discuss with the instructor how you can do better on following assignments.
Most recently updated on November 24, 2008.