No class Thursday April 15

Advanced AI concepts / Fundamentals of AI for Robotics 2004

(15-780 and 16-731, the Core Graduate AI course for CS and Robotics)

Instructor: Chris Atkeson,
Use AI (or just I) to generate email address. Hint: cga
Units: 12

Description:

This course is targeted at graduate students who need to learn about and use current research in Artificial Intelligence---the discipline of designing intelligent machines.

A hallmark of recent AI conference papers, journal papers and theses has been the incorporation of ideas from outside traditional AI. Techniques from Probability, Statistics, Economics, Algorithms, Operations Research and Optimal Control are increasingly important tools for improving the intelligence and autonomy of machines, whether those machines are robots surveying Antartica, schedulers moving billions of dollars of inventory, spacecraft deciding which experiments to perform, vehicles negotiating for lanes on the freeway, or data management systems that persistently scan for anomalies and trends. The primary content of this AI course is a review of a selected set of these tools. The course will cover the ideas underlying these tools, their implementation, and how to use them or extend them in your research.

Requirements and target audience: The course is tailored to participants who have already been exposed to some pre-PhD-level training in AI and programming. Students who feel they may be lacking these basics should consult the instructor, and may be advised to begin by taking 15-381, the undergraduate AI course.

Syllabus outline: We will travel through a wide range of general scenarios that might be encountered in the design of an intelligent systems such as embedded autonomous controllers, corporate logistics controllers or alarm and diagnosis systems. The scenarios are differentiated by various combinations of assumptions about

• Representation
• Search
• Adaptability/Learning
• Uncertainty

The best predictor for what this year's course will be like is to look at last year's course

Lectures (most of these are modified versions of Andrew Moore's lectures).

• Introduction

The first section of the course views AI as function approximation, with some probability reviewed.

• AI = Function Approximation (Part 1)
• AI = Function Approximation (Part 2)
• Review of Probability (Discrete stuff)
• Review of Probability (Continuous stuff)
• Review of Gaussians
• Decision Trees
• Memory-based Learning
• Support Vector Machines
• Cross Validation

This part of the course treats recognition and estimation over time from a probabilistic (Bayesian) viewpoint.

• Recognition and Estimation over Time
• Hidden Markov Models
• Speech Recognition
• Kalman Filter

Solving optimization problems

• Continuous Optimization
• Search
• Astar
• Constraint Satisfaction and Scheduling

Planning and Reinforcement Learning (Optimization over time)

• Markov Decision Processes: Discrete case
• Dynamic Optimization
• Reinforcement Learning
• Reinforcement Learning II
• Maze slides

Reasoning (mostly Bayesian)

• Bayesian reasoning: summary
• Bayesian reasoning: introduction
• Bayesian reasoning: inference
• Bayesian reasoning: structural learning

Clustering:

• Gaussian mixture models
• K means

Symbolic AI

Planning with others: Games, multiple agents (Optimization with other agents)

Assignment 1 (Due Tuesday Jan. 13):

• Send email to cga, neill, tingliu, and sonyaa@cs.cmu.edu with the following information:
• Tell us about yourself. Why are you taking this course?
• What do you know about AI? What course(s) have you taken? What did they cover? Feel free to point us to course web pages, or send syllabi.
• What research are you doing now? How does it relate to AI?
• What topics do you want to cover in this course? What topics should not be covered? Here is a likely list of topics we will cover. Here is a list of review keywords for AAAI 2004. Also check out textbooks, recent conferences, and journals. Ask your friends. Since it is impossible to cover everything in one semester, we will have to pick and choose.
• How can we help you?

Assignment 2 (SPAM assignment):
Due Feb 2, 2004, 11:59PM EST
Turn this in by sending email to cga@cmu.edu, sonyaa@cs.cmu.edu, neill@cs.cmu.edu and tingliu@cs.cmu.edu.

To get you started, I have collected some spam:

• almost-certainly-spam Email the Spam Assassin filter thinks is almost certainly spam.
• probably-spam Email the Spam Assassin filter thinks is probably spam.
• got-past-filter Email that I think is spam that either got through or bypassed the filter.

Assignment 3 (TRACKING assignment):
Due Feb 23, 2004, 11:59PM EST
Turn this in by sending email to cga@cmu.edu, sonyaa@cs.cmu.edu, neill@cs.cmu.edu and tingliu@cs.cmu.edu.

Here is what was turned in.

Assignment 4 (Due Thursday March 4):

• Send email to cga, neill, tingliu, and sonyaa@cs.cmu.edu with the following information:
• What are you going to do for your project?
• How can we help you?

Some suggested projects:

• Extend the Tracking Assignment in some way.
• Extend the Dynamic Motion Planning Assigment in some way.
• Do dynamic motion planning for a biped, especially for rough terrain.
• Process the video from class to identify body parts and gestures.
• Process the audio from class to produce an accurate transcript.
• Use the audio/video from class as a database to answer questions.
• Do something for Valerie (talk to Mark Michalowski his-last-name@cmu.edu). Here is a suggested list:
  • Speech recognition
  • Sound localization
  • Face detection or recognition (of pose or identity)
  • Natural language parsing and understanding
  • Different laser tracking method to locate people: particle/kalman filtering

Project FAQ:

• Can we work in groups? alone?
  Yes. The maximum group size is 3. You can work alone.
• Can we use stuff off the web?
  Yes. As long as you clearly indicate what your contribution is, using other resources is fine. You will be graded on the "value" you add to whatever resources you use.
• How do we turn this in?
  I would like a URL pointing to your writeup (and code), so we can make a class web page, and everyone can learn from what others do. Ideally, you can make your writeup available to the world, so others can build on what you do.

Assignment 5 (MAZE assignment):
Due April 19, 2004, 11:59PM EST
Turn this in by sending email to cga@cmu.edu, sonyaa@cs.cmu.edu, neill@cs.cmu.edu and tingliu@cs.cmu.edu.