CAREER: Runtime Recommender Systems for Compositional Modeling of Scientific Computations

点击翻译按钮获取中文摘要

ABSTRACTEIA 9984317Naren RamakrishnanVirginia Polytechnic Institute & State UniversityCAREER: Runtime Recommender Systems for Compositional Modeling of Scientific ComputationsThe central goal of this career development proposal is to introduce runtime recommendation, an abstraction that extends the above two ideas significantly. Specifically, it monitors a computational process, detects state-changes, and makes selections of solution components dynamically, thus aiding knowledge-based application composition at runtime. Such a facility is important in many problem domains because: (i) the nature of the problem being solved changes as the computations are being performed, (ii) the underlying computing platform or resource availability is dynamic, or (iii) information about application performance characteristics is acquired during the actual computation rather than before. While traditional recommenders are designed off-line (by organizing a battery of benchmark problems and algorithm executions, and subsequently mining it to obtain high-level recommendation rules), the design of a runtime recommender system is difficult, because such a database is not readily available and needs to be "captured" on the fly. Thus, a runtime recommender interacts dynamically with its environment and learns through interactions with its environment.