Opportunity at National Institute of Standards and Technology (NIST)
Applied Representational Learning
Information Technology Laboratory, Statistical Engineering Division
Please note: This Agency only participates in the February and August reviews.
|Lu, ZQ John
This project takes advantage of increasing activities at NIST on using AIs and machine learning in metrology and is aimed at addressing issues of what to measure and what inputs to use when data such as sensor signals, medical images, or chemical spectra are taken. In both prediction or classification problems, selecting what input variables or features based on high-dimensional measurement data is crucial for developing algorithms for prediction or classification (Bengio et al 2013) and we think that both transparency and subject matter understanding of the predictive features are equally important in analysis of complex and time-varying processes. The related issue of statistically principled use of metrics for evaluation is also important as demonstrated in diagnostic medicine (Zhou et al 2011). This opportunity will contribute to the algorithmic and prediction oriented approach to statistics which is currently going through a renaissance thanks to efforts by Leo Breiman (2001) and other pioneering statisticians, and increasing activities from outside statistics. The potential applications at NIST include fire prediction from sensors, anomaly detection in time series, and medical diagnosis from medical images such as optical coherence tomography or hyperspectral images.
Y. Bengio, A. Courville, P. Vincent (2013). "Representation Learning: A Review and New Perspectives". IEEE Transaction on Pattern Analysis and Machine Intelligence. 35: 1798–1828.
X-H Zhou , N A. Obuchowski, D K. McClish (2011). Statistical Methods in Diagnostic Medicine, Chapter 2, Wiley, 2nd edition. https://doi.org/10.1002/9780470906514.ch2.
Leo Breiman (2001). Statistical Modeling: The Two Cultures (with comments). Statistical Science Vol. 16, no. 3, 199--231.
Representational learning; feature learning; low-dimensional signals; image-based measurements; functional or spectral data; statistical prediction paradigm
Open to U.S. citizens
Open to Postdoctoral applicants