UCF AI-Forum

Home

Meetings

Next meeting

Friday, March 6, 2009, 3:00pm in HEC 101
Title: Overview of Evolutionary Computation Theory

Abstract:
During the last half century of Evolutionary Computation (EC) research, a multitude of algorithms have been developed that fit the very loose rubric of "inspired by natural evolution." Unsurprisingly, analytical approaches to understanding such systems are similarly diverse. Nevertheless, cursory study of EC typically provides some limited background in schema theory for genetic algorithms and successful real-world application of evolutionary algorithms (EAs) are more often justified analogically than by relating to underlying foundational aspects of EC. I believe this has led to a number of common myths about the state of theory for the field: that schema theory and the building block hypothesis is the only EC theory, that there is very little formal basis to these methods, or that EC theory is largely descriptive and has very little utility to practical application of EAs.

In fact, there have been a number of theoretical advances in EC. There are several critical conference and workshop venues that specifically target EC theory, and new advances are frequently published in major journals of the field---including the journal Theoretical Computer Science, perhaps the most prestigious journal in all of CS. EC theory incorporates mathematical tools from many disciplines (dynamical systems, probability theory, discrete mathematics, etc.) and considers many aspects of such systems (step-wise algorithm behavior, global algorithm performance, non-linear dynamics, challenges arising from relationships between problem properties and algorithm properties, etc.). The latest Foundations of Genetic Algorithms produced 18 papers covering analysis of evolution-inspired methods for continuous optimization, combinatorial optimization, multi-objective optimization, co-optimization, as well as problem analysis. None discussed schema theory and many offered predictive analysis of algorithm performance.

I offer a broad perspective in this talk. Like EC itself, EC theory is not at all unified, and I lay out a high-level, math-free organization categorizing existing work. At the top level, these include two different focuses of algorithm analysis (global and local behavior), component analysis and problem analysis. I discuss the famous No Free Lunch as a separate category. Within these categories, we find topics such as landscape-oriented Walsh analysis, exact/pressimisstic micro/maco schema theory, infinite population dynamical models, runtime analysis, and many more. I will select a handful of examples and provide math-light general discussions of them. The talk targets those with a basic understanding of computer science and optimization, assuming as little EC background knowledge as possible. Goals of the talk include creating a greater awareness of EC theory, providing a (hopefully) more accurate notion of extant strengths and weaknesses in current approaches and results, and stimulating interest in further discussion on these topics.

Speaker info:
This talk will be presented by Dr. R. Paul Wiegand of the Institute for Simulation & Training. He is the founder of UCF's NCCS Lab.

Download abstract: pdf

Previous meeting topics

Novelty search
Discussion of current research
Searle's Chinese Room argument
Joseph Weizenbaum
Symbolic and sub-symbolic representations
Computational and artificial creativity
Common sense reasoning
Marvin Minsky: "Steps Toward Artificial Intelligence"
Organizational meeting

Previous meeting details

Friday, February 13, 2009, 2:00pm in HEC 101
Title: Abandoning Objective and the Search for Novelty

Abstract:
This talk will present provocative results from recent work from the Evolutionary Complexity Research Group at UCF on a new algorithm called novelty search. The paradoxical idea in novelty search is that achieving an objective is sometimes easier if the objective is ignored. Instead of searching for the objective or (i.e. measuring fitness with respect to the objective), the algorithm simply searches for novelty. While the principle that objectives can be more effectively achieved by ignoring them is counterintuitive, it has proven true in a number of domains. This talk will focus on several such domains in which novelty search outperforms objective search and examine the implications for artificial intelligence and machine learning in general.

Speaker Bio:
Kenneth O. Stanley is an assistant professor in the School of Electrical Engineering and Computer Science at the University of Central Florida. He received a Ph.D. in 2004 from the University of Texas at Austin. He is the inventor of the NeuroEvolution of Augmenting Topologies (NEAT) algorithm for evolving complex artificial neural networks. His main research contributions are in neuroevolution (i.e. evolving neural networks), generative and developmental systems, coevolution, machine learning for video games, and interactive evolution. He has won best paper awards for his work on NEAT, NERO, NEAT Drummer, and HyperNEAT. He is the chair of the IEEE Task Force on Computational Intelligence and Video Games, and has chaired the Generative and Developmental Systems track at GECCO for the last three years.

Joel Lehman and Kenneth O. Stanley (2008). Exploiting Open-Endedness to Solve Problems Through the Search for Novelty. In the Proceedings of the Eleventh International Conference on Artificial Life (ALIFE XI). Cambridge, MA: MIT Press.
Download: pdf
Friday, October 3, 2008, 2:00pm in HEC 102
This month we will change course and discuss our own research interests and current projects. Here at UCF we have a number of students and professors researching various AI related topics such as artificial neural networks, natural language processing, evolutionary computation, machine learning, and multi-agent systems. This will be an informal meeting where everyone will have the opportunity to briefly discuss the problems, questions, and ideas that are of interest to them. The reading below is an inspiring talk given by Richard Hamming in 1986 about how to do great research. It's a great read for any scientist.

Richard W. Hamming (1986). You and Your Research. Transcription of the Bell Communications Research Colloquium Seminar.
Download: html pdf
Friday, September 12, 2008, 2:00pm in HEC 102
"Could a machine think?" This question will be central to our discussion in this month's meeting, as we debate John Searle's Chinese Room argument for why a machine cannot think. Searle classifies Artificial Intelligence research into two categories: "weak" or "cautious" AI and "strong" AI. With the weak AI view, the computer is a powerful tool that can be used to study the mind, whereas with the strong view of AI, the computer, if programmed appropriately, is a mind. Searle attacks the second viewpoint of AI in the paper given below, and his arguments will serve as a starting point for our discussion this month.

John R. Searle (1980). Minds, brains, and programs. Behavioral and Brain Sciences, 3(3): 417-457.
Download: html

Wikipedia's discussion on the Chinese Room argument
Download: html

Larry Hauser. Searle's Chinese Room Argument: Annotated Bibliography.
Download: html
Friday, April 4, 2008, 1:30pm in HEC 119
Joseph Weizenbaum was one of the original Artificial Intelligence researchers but later became one of AI's biggest critics. Weizenbaum gave us the ELIZA program in 1966, which was the first attempt to bridge the gap between the computer and human with a natural language interface. While Weizenbaum observed people interact with ELIZA as if it was an actual person, he became increasingly aware of the dangers of people's reliance on technology. This ended up leading Joseph Weizenbaum to denounce Artificial Intelligence research in favor of a more humanistic philosophy where computers don't make important decisions for us because they lack compassion and other human qualities. In this meeting we will discuss what led Dr. Weizenbaum to these ideas and if these are things we should keep in mind while doing our own research. The papers below will be a starting point for our discussion. The Drew McDermott paper was suggested by Weizenbaum as the most important paper in the AI literature (Interview with Weizenbaum).

Drew McDermott (1976), Artificial intelligence meets natural stupidity. ACM SIGART Bulletin, 57:4-9.
Download (on campus): pdf

Joseph Weizenbaum (1972), On the Impact of the Computer on Society. Science: New Series, 176(4035):609-614.
Download (on campus): pdf

Joseph Weizenbaum (1966), ELIZA - A Computer Program For the Study of Natural Language Communication Between Man And Machine. Communications of the ACM, 9(1):36-45.
Download (on campus): pdf

John Markoff (2008), Joseph Weizenbaum, Famed Programmer, Is Dead at 85. NY Times, March 13, 2008.
View: html

Diana ben-Aaron (1985), Weizenbaum examines computers and society. The Tech Online Edition, 105(16), April 9, 1985.
View: html
Friday, February 8, 2008, 1:30pm in HEC 101
The goal of this meeting is to define and understand symbolic and sub-symbolic knowledge representations. Both approaches have been used to model the human cognitive system and in the design of intelligent computer programs. Symbolic representations are based on symbol manipulation and formal logic, whereas sub-symbolic representations involve distributed representations such as artificial neural networks. We will concentrate on defining and understanding these concepts and will use next month's meeting to debate the use of the two in Artificial Intelligence research. The papers below can be used as a starting point for understanding the two approaches to knowledge representation and intelligence modelling.

Troy D. Kelley (2003), Symbolic and Sub-Symbolic Representations in Computational Models of Human Cognition. Theory & Psychology, 13(6):847-860.
Download (on campus): pdf

Randall Davis, Howard Shrobe, and Peter Szolovits (1993). What is a Knowledge Representation? AI Magazine, 14(1):17-33.
Download: pdf

David A. Medler (1998), A Brief History of Connectionism. Neural Computing Surveys (http://www.icsi.berkeley.edu/~jagota/NCS), 1(1).
Download: ps
Friday, January 18, 2008, 1:30pm in HEC 101
The focus of this meeting will be on Computational and Artificial Creativity, that is, the study of creativity from a computational perspective and the development of systems to emulate or at least support creativity. One of the most impressive traits of the human mind is its ability to cope with new problems by creating novel solutions from those it already possesses. How can a machine use the resources it is given to produce concepts novel to it and even to humans? What techniques are currently being used to do so? We will discuss these and related questions in this meeting.

Wlodzislaw Duch (2006), Computational Creativity. International Joint Conference on Neural Networks, pp. 435-442.
Download: pdf
Download (on campus): pdf

Israel Beniaminy (2007), Creativity - The Last Human Stronghold?. Online article, December 24, 2007.
View: link
Friday, November 2nd, 2007, 1:30pm in HEC 111
One thing that all AI researchers can agree upon is that the easy problems are hard, and the hard problems are easy. Common sense reasoning is one of the problems that's easy for humans but has proven difficult to solve computationally. Many researchers in the AI field have concentrated their efforts on accumulating large databases full of common sense knowledge. There has also been a lot of work that focuses on how best to represent this information. What is common sense knowledge, and why is it important to AI researchers? In this month's AI-Forum meeting, we will be discussing the answers to these questions and look at the research others have done with regards to these problems. The papers below will be used as a starting point for our discussion.

Douglas B. Lenat (1995), CYC: A Large-Scale Investment in Knowledge Infrastructure. Communications of the ACM, 38(11).
Download: pdf

Push Singh (2001). The Open Mind Common Sense project. KurzweilAI.net.
Download: pdf
For more information: Open Mind

Hugo Liu and Push Singh (2004). ConceptNet: A Practical Commonsense Reasoning Toolkit. BT Technology Journal, 22(4).
Download: pdf

Doug Lenat, George Miller, and Toshio Yokoi (1995). CYC, WordNet, and EDR: Critiques and Responses. Communications of the ACM, 38(11), pp. 45-48.
Download: pdf
Friday, October 5th, 2007, 1:00pm in HEC 101
In this meeting, we'll be discussing Marvin Minsky's 1961 paper, Steps Toward Artificial Intelligence. This early paper paved the way for future endeavors in Artificial Intelligence research and shows that the problems from 40 years ago are the same ones we're facing today. We welcome you to take part in this discussion.

Marvin Minsky (1961), Steps Toward Artificial Intelligence. Proc. IRE, 49(1), pp. 8-30.
Download: html pdf (from campus)
Friday, September 7th, 2007, 1:30pm in HEC 101
This was the first ever meeting of the AI-Forum. We discussed ideas about how to format future meetings. We also put together a student committee that will be responsible for future meetings.

UCF AI-Forum
Home	People	Meetings	Resources
Next meeting Friday, March 6, 2009, 3:00pm in HEC 101 Title: Overview of Evolutionary Computation Theory Abstract: During the last half century of Evolutionary Computation (EC) research, a multitude of algorithms have been developed that fit the very loose rubric of "inspired by natural evolution." Unsurprisingly, analytical approaches to understanding such systems are similarly diverse. Nevertheless, cursory study of EC typically provides some limited background in schema theory for genetic algorithms and successful real-world application of evolutionary algorithms (EAs) are more often justified analogically than by relating to underlying foundational aspects of EC. I believe this has led to a number of common myths about the state of theory for the field: that schema theory and the building block hypothesis is the only EC theory, that there is very little formal basis to these methods, or that EC theory is largely descriptive and has very little utility to practical application of EAs. In fact, there have been a number of theoretical advances in EC. There are several critical conference and workshop venues that specifically target EC theory, and new advances are frequently published in major journals of the field---including the journal Theoretical Computer Science, perhaps the most prestigious journal in all of CS. EC theory incorporates mathematical tools from many disciplines (dynamical systems, probability theory, discrete mathematics, etc.) and considers many aspects of such systems (step-wise algorithm behavior, global algorithm performance, non-linear dynamics, challenges arising from relationships between problem properties and algorithm properties, etc.). The latest Foundations of Genetic Algorithms produced 18 papers covering analysis of evolution-inspired methods for continuous optimization, combinatorial optimization, multi-objective optimization, co-optimization, as well as problem analysis. None discussed schema theory and many offered predictive analysis of algorithm performance. I offer a broad perspective in this talk. Like EC itself, EC theory is not at all unified, and I lay out a high-level, math-free organization categorizing existing work. At the top level, these include two different focuses of algorithm analysis (global and local behavior), component analysis and problem analysis. I discuss the famous No Free Lunch as a separate category. Within these categories, we find topics such as landscape-oriented Walsh analysis, exact/pressimisstic micro/maco schema theory, infinite population dynamical models, runtime analysis, and many more. I will select a handful of examples and provide math-light general discussions of them. The talk targets those with a basic understanding of computer science and optimization, assuming as little EC background knowledge as possible. Goals of the talk include creating a greater awareness of EC theory, providing a (hopefully) more accurate notion of extant strengths and weaknesses in current approaches and results, and stimulating interest in further discussion on these topics. Speaker info: This talk will be presented by Dr. R. Paul Wiegand of the Institute for Simulation & Training. He is the founder of UCF's NCCS Lab. Download abstract: pdf Previous meeting topics Novelty search Discussion of current research Searle's Chinese Room argument Joseph Weizenbaum Symbolic and sub-symbolic representations Computational and artificial creativity Common sense reasoning Marvin Minsky: "Steps Toward Artificial Intelligence" Organizational meeting Previous meeting details Friday, February 13, 2009, 2:00pm in HEC 101 Title: Abandoning Objective and the Search for Novelty Abstract: This talk will present provocative results from recent work from the Evolutionary Complexity Research Group at UCF on a new algorithm called novelty search. The paradoxical idea in novelty search is that achieving an objective is sometimes easier if the objective is ignored. Instead of searching for the objective or (i.e. measuring fitness with respect to the objective), the algorithm simply searches for novelty. While the principle that objectives can be more effectively achieved by ignoring them is counterintuitive, it has proven true in a number of domains. This talk will focus on several such domains in which novelty search outperforms objective search and examine the implications for artificial intelligence and machine learning in general. Speaker Bio: Kenneth O. Stanley is an assistant professor in the School of Electrical Engineering and Computer Science at the University of Central Florida. He received a Ph.D. in 2004 from the University of Texas at Austin. He is the inventor of the NeuroEvolution of Augmenting Topologies (NEAT) algorithm for evolving complex artificial neural networks. His main research contributions are in neuroevolution (i.e. evolving neural networks), generative and developmental systems, coevolution, machine learning for video games, and interactive evolution. He has won best paper awards for his work on NEAT, NERO, NEAT Drummer, and HyperNEAT. He is the chair of the IEEE Task Force on Computational Intelligence and Video Games, and has chaired the Generative and Developmental Systems track at GECCO for the last three years. Joel Lehman and Kenneth O. Stanley (2008). Exploiting Open-Endedness to Solve Problems Through the Search for Novelty. In the Proceedings of the Eleventh International Conference on Artificial Life (ALIFE XI). Cambridge, MA: MIT Press. Download: pdf Friday, October 3, 2008, 2:00pm in HEC 102 This month we will change course and discuss our own research interests and current projects. Here at UCF we have a number of students and professors researching various AI related topics such as artificial neural networks, natural language processing, evolutionary computation, machine learning, and multi-agent systems. This will be an informal meeting where everyone will have the opportunity to briefly discuss the problems, questions, and ideas that are of interest to them. The reading below is an inspiring talk given by Richard Hamming in 1986 about how to do great research. It's a great read for any scientist. Richard W. Hamming (1986). You and Your Research. Transcription of the Bell Communications Research Colloquium Seminar. Download: html pdf Friday, September 12, 2008, 2:00pm in HEC 102 "Could a machine think?" This question will be central to our discussion in this month's meeting, as we debate John Searle's Chinese Room argument for why a machine cannot think. Searle classifies Artificial Intelligence research into two categories: "weak" or "cautious" AI and "strong" AI. With the weak AI view, the computer is a powerful tool that can be used to study the mind, whereas with the strong view of AI, the computer, if programmed appropriately, is a mind. Searle attacks the second viewpoint of AI in the paper given below, and his arguments will serve as a starting point for our discussion this month. John R. Searle (1980). Minds, brains, and programs. Behavioral and Brain Sciences, 3(3): 417-457. Download: html Wikipedia's discussion on the Chinese Room argument Download: html Larry Hauser. Searle's Chinese Room Argument: Annotated Bibliography. Download: html Friday, April 4, 2008, 1:30pm in HEC 119 Joseph Weizenbaum was one of the original Artificial Intelligence researchers but later became one of AI's biggest critics. Weizenbaum gave us the ELIZA program in 1966, which was the first attempt to bridge the gap between the computer and human with a natural language interface. While Weizenbaum observed people interact with ELIZA as if it was an actual person, he became increasingly aware of the dangers of people's reliance on technology. This ended up leading Joseph Weizenbaum to denounce Artificial Intelligence research in favor of a more humanistic philosophy where computers don't make important decisions for us because they lack compassion and other human qualities. In this meeting we will discuss what led Dr. Weizenbaum to these ideas and if these are things we should keep in mind while doing our own research. The papers below will be a starting point for our discussion. The Drew McDermott paper was suggested by Weizenbaum as the most important paper in the AI literature (Interview with Weizenbaum). Drew McDermott (1976), Artificial intelligence meets natural stupidity. ACM SIGART Bulletin, 57:4-9. Download (on campus): pdf Joseph Weizenbaum (1972), On the Impact of the Computer on Society. Science: New Series, 176(4035):609-614. Download (on campus): pdf Joseph Weizenbaum (1966), ELIZA - A Computer Program For the Study of Natural Language Communication Between Man And Machine. Communications of the ACM, 9(1):36-45. Download (on campus): pdf John Markoff (2008), Joseph Weizenbaum, Famed Programmer, Is Dead at 85. NY Times, March 13, 2008. View: html Diana ben-Aaron (1985), Weizenbaum examines computers and society. The Tech Online Edition, 105(16), April 9, 1985. View: html Friday, February 8, 2008, 1:30pm in HEC 101 The goal of this meeting is to define and understand symbolic and sub-symbolic knowledge representations. Both approaches have been used to model the human cognitive system and in the design of intelligent computer programs. Symbolic representations are based on symbol manipulation and formal logic, whereas sub-symbolic representations involve distributed representations such as artificial neural networks. We will concentrate on defining and understanding these concepts and will use next month's meeting to debate the use of the two in Artificial Intelligence research. The papers below can be used as a starting point for understanding the two approaches to knowledge representation and intelligence modelling. Troy D. Kelley (2003), Symbolic and Sub-Symbolic Representations in Computational Models of Human Cognition. Theory & Psychology, 13(6):847-860. Download (on campus): pdf Randall Davis, Howard Shrobe, and Peter Szolovits (1993). What is a Knowledge Representation? AI Magazine, 14(1):17-33. Download: pdf David A. Medler (1998), A Brief History of Connectionism. Neural Computing Surveys (http://www.icsi.berkeley.edu/~jagota/NCS), 1(1). Download: ps Friday, January 18, 2008, 1:30pm in HEC 101 The focus of this meeting will be on Computational and Artificial Creativity, that is, the study of creativity from a computational perspective and the development of systems to emulate or at least support creativity. One of the most impressive traits of the human mind is its ability to cope with new problems by creating novel solutions from those it already possesses. How can a machine use the resources it is given to produce concepts novel to it and even to humans? What techniques are currently being used to do so? We will discuss these and related questions in this meeting. Wlodzislaw Duch (2006), Computational Creativity. International Joint Conference on Neural Networks, pp. 435-442. Download: pdf Download (on campus): pdf Israel Beniaminy (2007), Creativity - The Last Human Stronghold?. Online article, December 24, 2007. View: link Friday, November 2nd, 2007, 1:30pm in HEC 111 One thing that all AI researchers can agree upon is that the easy problems are hard, and the hard problems are easy. Common sense reasoning is one of the problems that's easy for humans but has proven difficult to solve computationally. Many researchers in the AI field have concentrated their efforts on accumulating large databases full of common sense knowledge. There has also been a lot of work that focuses on how best to represent this information. What is common sense knowledge, and why is it important to AI researchers? In this month's AI-Forum meeting, we will be discussing the answers to these questions and look at the research others have done with regards to these problems. The papers below will be used as a starting point for our discussion. Douglas B. Lenat (1995), CYC: A Large-Scale Investment in Knowledge Infrastructure. Communications of the ACM, 38(11). Download: pdf Push Singh (2001). The Open Mind Common Sense project. KurzweilAI.net. Download: pdf For more information: Open Mind Hugo Liu and Push Singh (2004). ConceptNet: A Practical Commonsense Reasoning Toolkit. BT Technology Journal, 22(4). Download: pdf Doug Lenat, George Miller, and Toshio Yokoi (1995). CYC, WordNet, and EDR: Critiques and Responses. Communications of the ACM, 38(11), pp. 45-48. Download: pdf Friday, October 5th, 2007, 1:00pm in HEC 101 In this meeting, we'll be discussing Marvin Minsky's 1961 paper, Steps Toward Artificial Intelligence. This early paper paved the way for future endeavors in Artificial Intelligence research and shows that the problems from 40 years ago are the same ones we're facing today. We welcome you to take part in this discussion. Marvin Minsky (1961), Steps Toward Artificial Intelligence. Proc. IRE, 49(1), pp. 8-30. Download: html pdf (from campus) Friday, September 7th, 2007, 1:30pm in HEC 101 This was the first ever meeting of the AI-Forum. We discussed ideas about how to format future meetings. We also put together a student committee that will be responsible for future meetings.