Ion Muslea

JAAMAS Journal 2026 Journal Article

Experiences Acquired in the Design of RoboCup Teams: A Comparison of Two Fielded Teams

• Stacy Marsella
• Milind Tambe
• Ion Muslea

Abstract Increasingly, multi-agent systems are being designed for a variety of complex, dynamic domains. Effective agent interactions in such domains raise some of the most fundamental research challenges for agent-based systems, in teamwork, multi-agent learning and agent modelling. The RoboCup research initiative, particularly the simulation league, has been proposed to pursue such multi-agent research challenges, using the common testbed of simulation soccer. Despite the significant popularity of RoboCup within the research community, general lessons have not often been extracted from participation in RoboCup. This is what we attempt to do here. We have fielded two teams, ISIS97 and ISIS98, in RoboCup competitions. These teams have been in the top four teams in these competitions. We compare the teams, and attempt to analyze and generalize the lessons learned. This analysis reveals several surprises, pointing out lessons for teamwork and for multi-agent learning.

JAAMAS Journal 2026 Journal Article

Hierarchical Wrapper Induction for Semistructured Information Sources

• Ion Muslea
• Steven Minton
• Craig A. Knoblock

Abstract With the tremendous amount of information that becomes available on the Web on a daily basis, the ability to quickly develop information agents has become a crucial problem. A vital component of any Web-based information agent is a set of wrappers that can extract the relevant data from semistructured information sources. Our novel approach to wrapper induction is based on the idea of hierarchical information extraction, which turns the hard problem of extracting data from an arbitrarily complex document into a series of simpler extraction tasks. We introduce an inductive algorithm, STALKER, that generates high accuracy extraction rules based on user-labeled training examples. Labeling the training data represents the major bottleneck in using wrapper induction techniques, and our experimental results show that STALKER requires up to two orders of magnitude fewer examples than other algorithms. Furthermore, STALKER can wrap information sources that could not be wrapped by existing inductive techniques.

AAAI Conference 2005 Conference Paper

Online Query Relaxation via Bayesian Causal Structures Discovery

• Ion Muslea

We introduce a novel algorithm, TOQR, for relaxing failed queries over databases, that is, over-constrained DNF queries that return an empty result. TOQR uses a small dataset to discover the implicit relationships among the domain attributes, and then it exploits this domain knowledge to relax the failed query. TOQR starts with a relaxed query that does not include any constraint, and it tries to add to it as many as possible of the original constraints or their relaxations. The order in which the constraints are added is derived from the domain’s causal structure, which is learned by applying the TAN algorithm to the small training dataset. Our experiments show that TOQR clearly outperforms other approaches: even when trained on a handful of examples, it successfully relaxes more that 97% of the failed queries; furthermore, TOQR’s relaxed queries are highly similar to the original failed query.

IJCAI Conference 2003 Conference Paper

Active Learning with Strong and Weak Views: A Case Study on Wrapper Induction

• Ion Muslea
• Steven N Minton
• Craig A. Knoblock

Multi-view learners reduce the need for labeled data by exploiting disjoint sub-sets of features (views), each of which is sufficient for learning. Such algorithms assume that each view is a strong view (i. e. , perfect learning is possible in each view). We extend the multi-view framework by introducing a novel algorithm, Aggressive Co-Testing, that exploits both strong and weak views; in a weak view, one can learn a concept that is strictly more general or specific than the target concept. Aggressive Co-Testing uses the weak views both for detecting the most informative examples in the domain and for improving the accuracy of the predictions. In a case study on 33 wrapper induction tasks, our algorithm requires significantly fewer labeled examples than existing state-of-the-art approaches.

ICML Conference 2002 Conference Paper

Active + Semi-supervised Learning = Robust Multi-View Learning

• Ion Muslea
• Steven Minton
• Craig A. Knoblock

ICML Conference 2002 Conference Paper

Adaptive View Validation: A First Step Towards Automatic View Detection

• Ion Muslea
• Steven Minton
• Craig A. Knoblock

AAAI Conference 2000 Short Paper

Selective Sampling with Co-Testing: Preliminary Results

• Ion Muslea
• and Craig A. Knoblock

We present a novel approach to selective sampling, co-testing, which can be applied to problems with redundant views (i.e., problems with multiple disjoint sets of attributes that can be used for learning). The main idea behind co-testing consists of selecting the queries among the unlabeled examples on which the existing views disagree.

AAAI Conference 2000 Conference Paper

Selective Sampling with Redundant Views

• Ion Muslea
• and Craig A. Knoblock

Selective sampling, a form of active learning, reduces the cost of labeling training data by asking only for the labels of the most informative unlabeled examples. We introduce a novel approach to selective sampling which we call co-testing. Co-testing can be applied to problems with redundant views (i. e. , problems with multiple disjoint sets of attributes that can be used for learning). We analyze the most general algorithm in the cotesting family, naive co-testing, which can be used with virtually any type of learner. Naive co-testing simply selects at random an example on which the existing views disagree. We applied our algorithm to a variety of domains, including three real-world problems: wrapper induction, Web page classification, and discourse trees parsing. The empirical results show that besides reducing the number of labeled examples, naive co-testing may also boost the classification accuracy.

AAAI Conference 1999 Short Paper

Active Learning for Hierarchical Wrapper Induction

• Ion Muslea
• Steve Minton
• Craig Knoblock
• University of Southern California

As an alternative to manually writing extraction rules, we created STALKER, which is a wrapper induction algorithm that learns high-accuracy extraction rules. The major novelty introduced by STALKER is the concept of hierarchical wrapper induction: the extraction of the relevant data is performed in a hierarchical manner based on the embedded catalog tree (ECT), which is a user-provided description of the information to be extracted.

AIJ Journal 1999 Journal Article

Building agent teams using an explicit teamwork model and learning

• Milind Tambe
• Jafar Adibi
• Yaser Al-Onaizan
• Ali Erdem
• Gal A. Kaminka
• Stacy C. Marsella
• Ion Muslea

Multi-agent collaboration or teamwork and learning are two critical research challenges in a large number of multi-agent applications. These research challenges are highlighted in RoboCup, an international project focused on robotic and synthetic soccer as a common testbed for research in multi-agent systems. This article describes our approach to address these challenges, based on a team of soccer-playing agents built for the simulation league of RoboCup—the most popular of the RoboCup leagues so far. To address the challenge of teamwork, we investigate a novel approach based on the (re)use of a domain-independent, explicit model of teamwork, an explicitly represented hierarchy of team plans and goals, and a team organization hierarchy based on roles and role-relationships. This general approach to teamwork, shown to be applicable in other domains beyond RoboCup, both reduces development time and improves teamwork flexibility. We also demonstrate the application of off-line and on-line learning to improve and specialize agents' individual skills in RoboCup. These capabilities enabled our soccer-playing team, ISIS, to successfully participate in the first international RoboCup soccer tournament (RoboCup'97) held in Nagoya, Japan, in August 1997. ISIS won the third-place prize in over 30 teams that participated in the simulation league.

IJCAI Conference 1999 Conference Paper

Two Fielded Teams and Two Experts: A RoboCup Challenge Response from the Trenches

• Milind Tambe
• Gal A. Kaminka
• Stacy Marsella
• Ion Muslea
• Taylor Raines

The RoboCup (robot world-cup soccer) effort, initiated to stimulate research in multi-agents and robotics, has blossomed into a significant effort of international proportions. RoboCup is simultaneously a fundamental research effort and a set of competitions for testing research ideas. At IJ- CAI'97, a broad research challenge was issued for the RoboCup synthetic agents, covering areas of multi-agent learning, teamwork and agent modeling. This paper outlines our attack on the entire breadth of the RoboCup research challenge, on all of its categories, in the form of two fielded, contrasting RoboCup teams, and two off-line soccer analysis agents. We compare the teams and the agents to generalize the lessons learned in learning, teamwork and agent modeling.