Jordan B. Pollack - US grants

9512448 Mataric The proposed project is to establish and support a new laboratory for adaptive behavior which will be equipped with mobile robots; and the computational, electronic, and mechanical equipment. The laboratory supports interdisciplinary research involving engineering and cognitive science in behavior and learning on autonomous agents. Studies to be carried in areas of directed and emergent programming, reinforcement learning, neural network learning, and evolutionary and genetic programming approaches. A set of related research projects in multi-agent and multi-robot will be initiated for dealing with a variety of agents, environments, and tasks; for selecting control strategies; and for simplifying the task of designing complex multi-agent systems. The laboratory will also be used for student projects in advanced computer science courses. ***

The Recursive Auto Associative Memory (RAAM) architecture for compositional neural encoding has been the basis for numerous studies and prototype models for neural network algorithms applied to linguistic and symbolic reasoning tasks. However, to date, the scale and understanding of the representational model has been very limited. This is due to several logical conundrums inherent in the original model, related to the separation of terminal from non-terminal patterns. A deeper understanding of the logical structure of recursive encoding, and a new mathematical understanding of the fractal limit dynamics of recurrent networks, results in a novel and powerful revision of the RAAM architecture which resolve the logical problems and allows for potentially infinite structures to be cleanly and precisely represented by fixed-dimensional neural activity patterns. In this grant we will be exploring, refining, and exploiting this new architecture and building larger-scale models, using both serial and massively parallel implementations, which will increase the applicability of neural network learning systems to areas with stronger data structure requirements. Scientifically, this work builds strong connections between traditional cognitive science capacities, nonlinear dynamics, and neurally plausible computational models.

This SGER is a preliminary grant to begin testing a hypothesis concerning using the web for learning guided by the use of Multi-User Virtual Environments as a place for competitive and collaborative learning games. The hypothesis that this award is testing is that by tracking user performance, managing the set of available playmates for every student, and introducing virtual agent players at a variety of school levels, such a community of evolving learners can keep all participants appropriately challenged and motivated to learn. To test this hypothesis, appropriate games and learning outcomes will be defined and the software protocols written to do playmate matching agent choice, and collection of data or student learning. The students tested will be in fourth and fifth grade classrooms and work with games to improve learning of mathematics and elementary physics.

Title: ITR Award: Creating One to One Learning Opportunities across the Internet

We will study a new kind of "anytime" learning system, based on one to one peer interactions shielded by and facilitated on the Internet. Our pedagogical hypothesis is that students learn best when continually challenged in a diverse environment. Matching students with appropriate learning partners and curricular material can be done with today's Internet technology and peer to peer software. Existing technologies from machine learning and game theory - specifically those related to mediocrity as an equilibrium phenomenon in self organizing systems - will be implemented as practical technologies that can adapt automatically as students learn. The technology is based on simple techniques to bring humans together to learn, rather than domain-intensive knowledge-based tutoring of a single human, and can work with the existing technology infrastructure of most public schools. Internet multi-player gaming environments have been proven to scale to millions of users, as well as having high intrinsic motivation. The research incorporates fundamental research in Information Technology, sound experimental research on motivation in multi-player games, and a novel and scaleable distributed community for learning.

This project expands the capability of man-and-machine design through a creativity enhancing tool which amplifies the human design process by a novel fusion of interactive design with generative coevolution. The project involves the development of a Web-based, Interactive Evolutionary Algorithm to enhance the creativity of human users in designing complex three-dimensional forms, such as tensegrity architecture. A Co-Evolutionary User Modeling Algorithm capable of generating and maintaining a population of user models will serve as critic and collaborator for the human user. By learning the user's implicit goals and cognitive style we can accelerate the creative design process. A creativity-enhancing generative representation capable of encoding families of designs will capture underlying design dimensions. The design engine will be responsible for integrating the information from the evolved user models in order to automatically generate both surprising (critical) and predictable (collaborative) candidate designs for the human designer. The project develops measures of the creativity of the human-computer design system as a whole, by measuring the complexity and novelty of designs produced by the human user. The ability to enhance a user's creativity, and the ability to rapidly synthesize the demands of a large set of users, has immediate applications in several design domains outside of engineering and architecture. The enhancement of educational software, the design of consumer packaging, and artistic endeavors, among many others, all have much to gain from an open ended framework capable of collaborative discovery.