This project is encouraging interdisciplinary machine learning research and education that integrates the use of machine learning into the application areas, solves real science problems, and serves as a catalyst for new research in both machine learning and the application domains. We are investigating four basic machine learning issues that arise when working with real-world applications: (1) optimizing over choices available when generating training data; (2) assessing and improving the quality of training data; (3) designing specific algorithms and methods for time series and feature-based data; and (4) developing methods for abstaining during classification. Our research is motivated by on-going collaborations with researchers to create solutions for: training an artificial nose (Chemistry, Tufts); land-cover mapping from remotely sensed data (Geography, Boston University); classification of sky surveys (Astronomy, Harvard); non-invasive gluclose monitoring (Biomedical Engineering, Tufts); and liquification prediction (Civil Engineering, Tufts). The successful application of machine learning to each of the five tasks will have significant impact on the lives of humans. Our education initiatives have two complementary goals: (1) to educate computer science students on how to conduct interdisciplinary machine learning research; and (2) to educate professors, graduate students and undergraduates from science, engineering and medicine on how to recognize and pose problems as machine learning and data mining problems.
