When working with complex phenomena, or incomplete data, making predictions can be dominated by uncertainty. It is often unclear what predictive models are best and whether they are ultimately good enough. We have employed frequentist and computational Bayesian techniques to estimate uncertainty levels in statistical models, thereby clarifying the best model from several candidate solutions that otherwise seemed equal. In other cases, we have used anomaly detection approaches, such as outlier detection via random sample consensus (RANSAC). Our approach to uncertainty quantification leads to better predictions, a more robust understanding of the problem at hand, and ultimately delivers solutions that not only give an answer, but quantify how good that answer is.

Many engineering problems involve data that changes over time. We are well-versed in trend analysis, time-marching transient modeling, and time series analysis. Our team of data scientists and engineers understands the specific nuances of time series data, such as serial correlation, data filtration, stationarity, and spectral analysis. We have evaluated photon Doppler velocimetry (PDV) signals using Short-Time Fourier Transform (STFT) spectral analysis, Short-Time Auto Regression (STAR) spectral analysis, Hilbert transformations, etc. We routinely construct and apply data filters and differencing methods, and have also developed pattern-based feature extraction methods to reduce the need for subjective analyst interpretation.