Intelligent Systems (EECE 7266/8266)
Fall 2012
Instructor: Bonny Banerjee, Ph.D.
Contact Information:
Office: 208B Engineering Science Bldg
Phone: 901-678-4498
E-mail: BBnerjee@memphis.edu
Office Hours: MWF 1:30-2:00pm
When: MWF 11:30am-12:25pm
Where: Engineering Science Bldg. Room 222
Note: In Fall 2012, this course is being offered as "Intelligent Systems." Even though it is listed in the catalog as "Prolog Processing for Intelligent Systems," there will not be anything to do with Prolog. Unlike what is stated in the catalog, there is no prerequisite for this course. Also, this course is suitable for students with different backgrounds, including but not limited to, Engineering, Computer Science, Psychology, Philosophy, and Mathematics. This webpage supersedes any previous webpage for the course. Talks by invited speakers on different intelligent systems are planned during the latter half of the course. This webpage will be updated with the schedule as they are confirmed. Everyone, including faculty, is welcome to attend the talks. [Please scroll down for the list of external speakers.]
Course Description:
In this course, we will concentrate on three fundamental issues – knowledge acquisition, knowledge representation, and knowledge manipulation – for building intelligent systems. We will study perception and learning in visual, auditory and artificial modalities, and their implementation in artificial systems. We will study the different kinds of knowledge representations in an artificial system, such as, logical representations, template-based representations, probabilistic representations, neural representations, etc. We will study the different knowledge manipulation techniques in an artificial system, such as, logical reasoning, probabilistic reasoning, etc. and compare their strengths and weaknesses. Finally, we will discuss a few well-known intelligent systems some of which have performed at par with humans in the task for which they were built while some of which have failed, and what can be learned from their successes and failures.
Required Text:
"Artificial Intelligence: A Modern Approach" by Stuart Russell and Peter Norvig
Readings from research papers and different book chapters
Topics and tentative schedule (15 weeks):
Week 1 (08/27/12). What is “intelligence”?
Course aims and agenda
History of Artificial Intelligence: Brain-computer analogy, Chinese Room argument, Turing test, Fathers of AI and their contributions, Traditional AI approaches and their shortcomings (Don't miss the interesting debates here)
Week 2 (09/03/12). Generality and efficiency in the design of intelligent agents
Generality: No Free Lunch theorem (Wolpert & Macready, 1996; Ho & Pepyne, 2002), Common Cortical Algorithm hypothesis (e.g., von Melchner et al., 2000)
Efficiency: Complexity Theory
Intelligent agents
(Project proposals due by email by 09/03/12 before class)
Week 3 (09/10/12). Problem solving, Representation and Reasoning
Problem solving and search
Predicate logic, First-order logic
Logical inference: Deduction, Induction, Abduction (Josephson & Josephson, 1996)
Week 4 (09/17/12). Problem solving, Representation and Reasoning
Frames, Schemas, Cognitive maps, Embodied cognition (“Intelligence without representation” – Brooks)
Decision trees, Bayesian networks
Cognitive Architectures: Soar (Laird, 2008), ACT-R (Also see reviews on Cognitive Architectures research, e.g. Chong et al., 2007; Duch et al., 2008; Langley et al., 2008)
Week 5 (09/24/12). Neural networks
Perceptron, Multilayered perceptrons
Hopfield's auto-associative memory (learning by energy minimization)
Week 6 (10/01/12). Learning
Clustering (unsupervised): k-means clustering, Kohonen's self-organizing map (Also see review of clustering algorithms, e.g. Jain et al., 1999)
Classification (supervised): gradient descent, stochastic gradient descent, backpropagation algorithm (Also see review of approaches to classification,
e.g. Kotsiantis, 2007)
Weeks 7-8 (10/08/12). Preliminary presentations
(10/13/12-10/16/12 Fall break)
Week 9 (10/22/12). Visual Perception
History of object recognition research
Hierarchical feature learning (See a review in Bengio, 2009)
Visual routines (Ullman, 1996), Learning invariances (Rao, 1998)
Week 10 (10/29/12). Auditory Perception and Perception in artificial modalities
Hidden Markov Models, Hierarchical Hidden Markov Models
Speech recognition using HMM and MFCC, Learning hierarchy of speech features
Artificial soft and hard sensors (e.g. ad clicks, sonar) and applications
Week 11 (11/05/12). Intelligent Surveillance and Transportation Systems
Intelligent surveillance in public places, Intelligent traffic surveillance (Coifman & Banerjee, 2002), Autonomous parking and driving systems
Week 12 (11/12/12). Intelligent Tutoring Systems
Different tutoring systems developed at our Institute for Intelligent Systems (See many of them here)
11/12/12 Dr. Andrew Olney: Building the Guru Intelligent Tutoring System
11/14/12 Blair Lehman: Intelligent Tutoring Systems that Adapt to Student Cognition and Emotion
Week 13 (11/19/12). Robotics
Intelligent robotic applications in the industry and household (Don't forget to watch the beautiful movie A.I. at your leisure)
11/19/12 Dr. Mark H. Myers: Experimental Design for EEG Based Brain Computer Interface (BCI) to Model Interpersonal Coordination Dynamics
11/21/12 Exam
(11/23/12 Thanksgiving holiday)
Week 14 (11/26/12). Holy Grails and Controversies in developing Intelligent Systems
Holy grails of AI: A common substrate for low-level perception and high-level cognition, Problem of representation change (Fink’s system, Push’s system)
Are these truly intelligent systems (e.g., IBM’s Deep Blue)?
Necessary functions of an intelligent system
Week 15 (12/03/12). Final presentations
(Final project reports due by email by 12/05/12 before class)
Evaluation and Final Grades:
This course requires high level of creative, research-oriented activities from each student. Grading will include the following components: midterm and final exams (20%+20%), a paper presentation (10%), project including report (40%), class participation (10%). Students will develop a comprehensive project in their selected topic, including theoretical foundations, implementation (in any language of the student's choice), and testing using publicly available data or simulated environment. The instructor will suggest projects in class that may be pursued solo or in small groups. Any interesting project from the students will be very much welcome.
Each student will anonymously grade all the other students in the class based on the two presentations. The grade should contain a brief writeup of what was interesting in the project and suggestions for improvement (akin to a journal review). The quality of this grading by each student, as judged by the instructor, will constitute his 10% class participation. The 40% for the project will consist of 30% of the instructor's grade and 10% of the grade received from all the other students.
The 7266 and 8266 sections will be graded separately. In each exam, the students enrolled for 8266 will have to answer one more question. Also, the paper presentation is only for 8266 students; that 10% for 7266 students will be calculated from their final exam score.
This course is for those who cherish to let their imaginations go wild and create something extraordinary out of it!
External Speakers: (will be updated as confirmed)
11/12/12 11:30am-12:25pm: Dr. Andrew Olney, Dept. of Psychology and Institute for Intelligent Systems, University of Memphis
Title: Building the Guru Intelligent Tutoring System
Abstract: Intelligent Tutoring Systems are artificially intelligent computer programs that adapt to students' knowledge level in order to teach them efficiently. Typical development of ITSs focus on this core detail and assume a particular pedagogical theory to define the teaching component. In this talk, I will describe another approach
to this problem which is data-driven. In the Guru project, we videotaped, transcribed, coded, and data-mined tutoring strategies from authentic human sessions. AI and crowd-sourcing techniques were used to develop the tutoring system and content based on these models.
11/14/12 11:30am-12:25pm: Blair Lehman, Dept. of Psychology and Institute for Intelligent Systems, University of Memphis
Title: Intelligent Tutoring Systems that Adapt to Student Cognition and Emotion
Abstract: Intelligent Tutoring Systems were initially developed to provide adaptive scaffolding and instruction based on students' cognitive states. Recent research, however, has shown that to maximize student learning, Intelligent Tutoring Systems must also adapt to student emotions. This line of research has sought to answer three main questions: (1) what emotions occur during learning? (2) how can emotions be detected and tracked during learning? and (3) how should an intelligent tutoring system adaptively respond to student emotions to maximize learning? These three questions have been investigated with AutoTutor, an intelligent tutor system that hold mixed-initiative dialogues with students. Major findings from research on emotions during learning with AutoTutor along with future research directions will be discussed.
11/19/12 11:30am-12:25pm: Dr. Mark H. Myers, Hamilton Eye Institute, University of Tennessee Health Science Center
Title: Experimental Design for EEG Based Brain Computer Interface (BCI) to Model Interpersonal Coordination Dynamics
Abstract: Studies in EEG analysis from human subjects have demonstrated that beta oscillations carried perceptual information in spatial patterns across the cortex featuring amplitude and phase modulation occurrences when the subjects were engaged in task oriented activities. A hypothesis was tested that similar patterns could be found in the scalp EEG of human subjects during visual stimulation. Spatial patterns of EEG phase modulation were found that could be classified with respect to stimulus. Our results suggest that the scalp EEG can yield information about the timing of episodically synchronized brain activity in higher cognitive function, so as to support mechanisms of brain–computer interfacing.