High-Performance CPU-GPU Compute Cluster for Research on Computational Game Theory and Biological Steering
Abstract
We propose to purchase a high-performance compute cluster for hybrid CPU-GPU computing for research on computational game theory and biological steering in Prof. Tuomas SandholmÕs research group at Carnegie Mellon UniversityÕs Computer Science Department. This research and research-related education has strong relevance to the DoD, both in terms of existing academic DoD research grants and in technology transition to the DoD. On the biology steering side, the vision is the following. Living organisms adapt to challenges through evolution and adaptation. These survival mechanisms have proven to be a key difficulty in developing therapies, since the challenged organisms develop resistance. It would be desirable to be able to harness evolution and adaptation for therapeutic, technological, and scientific goals. For example, through a sequence of appropriate manipulations, could we get a heterogeneous population of cancer cells to evolve to benign ones? Or, could we steer the evolution of the population to a state where we can destroy it? Could we evolve bacteria that eat toxins from the environment? OrÑas the topic of our current ARO grant that is the first project along this new visionÑcan we steer a personÕs own T-cell population to a state that causes the immune system to better treat the disease at hand, for example, infections, cancers, and autoimmune diseases? To accomplish this, in 2012 I introduced the wild idea of steering evolution/adaptation strategically using computational game theory and opponent exploitation techniques. In this paradigm, a sequential contingency plan for steering evolution/adaptation is constructed computationally for the setting at hand. We are pursuing three different approaches to the problem: game-theoretic equilibrium finding in a game between the treater and the disease, exploiting the biological opponent by deep reinforcement learning and other algorithms, and exploiting the biological opponentÕs myopia. On the ARO project we have already published advances along each of these three avenues, but we are bottlenecked by the lack of adequate computational resources. These algorithms require significantly more CPU computing than my laboratoryÕs current 9-year-old compute cluster has. Also, some of these algorithms are best suited for hybrid CPU-GPU computing, and my current cluster does not have GPUs and GPUs cannot be attached to that old technology. The proposed new cluster will enable the research to progress and accelerate. On the computational game theory side, we have become a world leader, and are continuing an active research agenda that ranges from modeling to automated abstraction to equilibrium-finding techniques to algorithms for equilibrium refinements, and beyond. This has applications not only in the biological steering domain discussed above but also in a variety of other domains. As an important broad class of domains, most DoD planning, re-planning, wargaming, and training settings are really games because one needs to plan in the face of an adversary. Yet they are not treated as games today, which leads to plans that are frail and seem significantly better than they are. The game-theoretic methodology solves that problem. The research is about making it increasingly scalable, general, and sophisticated. My current compute cluster is wildly inadequate for these computations that require significantly more CPU computing and some of them also require large-scale hybrid CPU-GPU computing. The proposed new cluster will solve this problem. We have a proven technology transfer path for the fruits of this research to benefit the DoD. The proposed cluster will also have significant research-related education value. My PhD students and undergraduate students working on these research projects will use the cluster actively.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 09, 2020
- Source ID
- W911NF2010081
Entities
People
- Tuomas Sandholm
Organizations
- Army Contracting Command
- Massachusetts Institute of Technology
- United States Army