Join Brian Rohrback for his talk on ChemMLometrics. Perform. Repeat., November 16th at 10:15AM.

ChemMLometrics. Perform. Repeat.

Doing a one-off research project can be very satisfying, but in industry the money is made when the results of such a project can be placed into routine quality assessment.  There is much discussion today revolving around machine learning and some of this discussion has invaded the space that has been occupied by chemometrics over the last few decades. There is a large overlap as these techniques are applied in instruments aimed at process quality control.  To a large extent, any differentiation will simply be a question of the jargon chosen. An on-line Stanford University class on machine learning covers Principal Components Analysis, Principal Components Regression, Partial Least Squares, K-Nearest Neighbor, and other mainstays of the chemometrics toolbox.  However, there are techniques that are new in the machine learning realm that can be employed for targeted tasks, improving the traditional chemometrics regime of supervised learning.  The key is to narrow the focus for each component in a system and to understand the extent to which you can control the inputs.  This talk highlights examples in optical spectroscopy calibration particularly for the identification of outliers, and in chromatography for both signal processing and the management of chromatographic libraries. The emphasis is to highlight both the similarities and the uniqueness of machine learning and chemometrics and show how they combine to form robust solutions for industry.

Join Brian Rohrback for his talk on Automating Calibrations for Optical Spectroscopy, November 16th at 1:30PM.

Automating Calibrations for Optical Spectroscopy

This talk represents a summary of a multi-industry consortium established eight years ago to re-evaluate how the calibration process in optical spectroscopy could be managed more efficiently.  The idea is to enable a shift from current practices to approaches that take better advantage of the computational power and some newer concepts supplied by research into machine learning algorithms.  The result is a solution that is not disruptive of any legacy instruments or software already in place and lessens the workload rather than laying on additional requirements. The approach uses all readily available components and can be assembled easily for any specified purpose.  The use of commercial components reduces the cost of deployment and assembling pieces in a plug-and-play manner minimizes the impact of any previous selections of hardware and software.