next day
all days

View: session overviewtalk overviewside by side with other conferences

09:30-10:30 Session 1
EAAI-13 Invited Talk: Learning in the Lab at Midnight: Experiences from Teaching AI at Berkeley and Online
SPEAKER: unknown

ABSTRACT. Where does learning really happen? Only a little happens in lecture; most students learn much more working with friends in the lab at midnight. The modern student experience increasingly revolves around coursework, peer assistance, and asynchronous interactions — not lectures, textbooks, and office hours. With these trends only increasing as enrollments rise and online channels emerge, how should we design our AI courses?

I'll talk about the best answers we've found so far for the Berkeley AI course. One key component of our approach is a set of thematically coherent, autograded projects that engage students and integrate with lectures in an ongoing way. More generally, I'll focus on several questions that have shaped our course, including: What should the role of a modern lecture be? What's the balance between cooperative learning and competition? When is an autograder more useful than a human TA? Why are students even taking AI in the first place? Finally, I'll talk about how technology that we originally developed for pedagogical purposes, such as rich autograding, has helped the course scale from tens to hundreds of students on campus and now to tens of thousands online.

Our experiences have resulted in a large number of re-usable materials, which we're always excited to share. I'll conclude with a discussion of how other instructors can take advantage of our lectures, interactive assignments, and autograded projects, which have already been used by over a hundred AI courses.

10:30-11:00Coffee Break
11:00-11:45 Session 2
Learning about Representational Modality: Design and Programming Projects for Knowledge-Based AI
SPEAKER: unknown

ABSTRACT. Many AI courses include design and programming projects that provide students with opportunities for experiential learning. Design and programming projects in courses on knowledge-based AI typically explore topics in knowledge, memory, reasoning, and learning. Traditional AI curricula, however, seldom highlight issues of modality of representations, often focusing on propositional representations. In this paper, we report on an investigation into learning about representational modality through a series of projects based around geometric analogy problems similar to the Raven's Progressive Matrices test of intelligence. We conducted this experiment over three years, from Fall 2010 through Fall 2012, in a class on knowledge-based AI. We used the methodology of action research in which the teacher is also the researcher. We discovered that students found these projects motivating, engaging, and challenging, in several cases investing significant time and posting their work online. From our perspective, the projects accomplished the goal of learning about representational modality in addition to knowledge representation and reasoning.

SEPIA: A Scalable Game Environment for Artificial Intelligence Teaching and Research
SPEAKER: unknown

ABSTRACT. We describe a game environment we have developed that we call the Strategy Engine for Programming Intelligent Agents (SEPIA). SEPIA is based on real-time strategy games, but modified extensively to preferentially support the development of artificial agents rather than human play. Through flexible configuration options, SEPIA is designed to be pedagogically scalable: suitable for use at the undergraduate and graduate levels, and also as a research testbed. We also describe assignments and our experiences with this environment in undergraduate and graduate classes.

11:45-13:45Lunch Break
13:45-14:30 Session 3
Model AI Assignment: Recreating TD-Gammon
SPEAKER: unknown

ABSTRACT. In this programming assignment, students in a junior/senior level undergraduate AI class reconstruct the classic TD-Gammon program from the original specifications in [Tesauro 1995]. We provide them with infrastructure code in Java for the backgammon player and the neural net evaluation function. The assignment builds on a prior assignment in which students build a hill-climbing learner that tunes a parametric evaluation function. The assignment specifies the TD algorithm in detail and identifies key design choices that have to be made. Students empirically determine a good set of design parameters for the algorithm. The assignment also has an analytical component in which students compare and explain the performance of the TD learner relative to the hill climbing learner. They are also invited to discuss the utility of this learning approach for other stochastic games.

Model AI Assignment: An Introduction to Counterfactual Regret Minimization
SPEAKER: unknown

ABSTRACT. In 2000, Hart and Mas-Colell introduced the important game-theoretic algorithm of regret matching. Players reach equilibrium play by tracking regrets for past plays, making future plays that are proportional to positive cumulative regrets (i.e. how much they wished they had made the moves on average). The technique is not only simple and intuitive; it inspired the counterfactual regret minimization algorithm that has since formed the basis for bots dominating annual computer poker competitions. Since these algorithms are relatively recent, there are few curricular materials available to introduce students, researchers, and practitioners to regret-based algorithms. The tutorial PDF, suggested exercises, and sample code offered below represent a modest first step towards making such recent innovations more accessible. Using Java code examples in Donald Knuth's literate programming style, we will present example applications of the regret matching algorithm for normal form "one-shot" games, counterfactual regret minimization (CFR) for extensive form games, and fixed-strategy iteration counterfactual regret minimization (FSICFR). We also briefly touch on strategy cleaning and how one might take maximal advantage of opponent error. Throughout, the reader first sees and example application and is then invited to deepen understanding through application to additional challenge problems.

Model AI Assignment: Tsunami Warning System: A Case Study of Intelligent Agents in Action
SPEAKER: unknown

ABSTRACT. This case study analyzes a tsunami warning system (TWS) from the perspective of an intelligent agent, the Tsunami Activity Reporter. Students receive a description of the reporter and an overview of a warning system from the National Oceanic and Atmospheric Administration and design an intelligent agent using the PEAS (performance measure, environment, actuators, and sensors) framework. Students independently complete a draft of the assignment then collaborate in small groups on a robust solution. TWS reinforces concepts of agent architecture and critical thinking skills while presenting a real-world and familiar context for problem solving. It presents an opportunity to invite in experts from other disciplines and an opportunity to discuss the ethics involved in warning systems and the dangers of false negatives and false positives. Experience indicates that as students debate the capabilities of various architectures and apply their knowledge of environments, they gain a deeper understanding of issues involved in working with intelligent agents. TWS is designed for Introduction to AI students but is easily adapted. K-12/CS1/CS2 students can learn the potential and challenges of intelligent systems. Students studying knowledge-based systems can implement prototype warning systems. Systems engineering students can consider interfaces between various sensors and software systems. Model AI Assignments AI applications Multi-agent systems Pfaffmann Jeff abstain abstain Levy Simon abstain abstain 7 Poole David abstain abstain 6 abstain abstain 271 Levy Simon 7 FALSE FALSE FALSE 274 Pfaffmann Jeff 7 FALSE FALSE FALSE 275 Poole David 6 FALSE FALSE FALSE

14:30-15:30 Session 4
15:30-16:00Coffee Break
16:00-17:00 Session 5
Poster: "Paradigms of AI Programming" in Python
SPEAKER: unknown

ABSTRACT. Norvig's (1992) Paradigms of AI Programming is an important book for learning about AI programming. However, the book uses Common Lisp as the programming language, which is less popular now than in 1992. Thus, we have translated many classical AI programs described in the book into Python, a more commonly used language. We have also documented the programs and offered them as a resource in a course on knowledge-based AI.

Poster: Student-Friendly Java-Based Multiagent Event Handling
SPEAKER: unknown
17:30-18:30 Session 6: EAAI-13 Social Hour, Hyatt Regency Hotel Lobby