FTC 2018 Short Courses
Short courses will be held on October 3, 2018.
Bayesian Statistics for Better Process Understanding & Prediction
Katherine Giacoletti, SynoloStats LLC
This course will provide an introduction to Bayesian statistical methodology for the applied statistician and, via examples and case studies, demonstrate how the use of such methods provides more precise process modeling and prediction than traditional analyses. The purpose of the course is not to provide instruction in programming, software use, or detailed computation, though it will include an overview of Bayesian methodology and some discussion of programming options. The learning objective is instead to provide the practitioner with a better understanding of the applicability of Bayesian statistics to manufacturing processes. Specifically, attendees will gain understanding to enable them to:
– Identify situations in which Bayesian methods may be of particular benefit
– Interpret the results (know how interpretation is different using Bayesian statistics as opposed to frequentist methods)
– Understand the implications of Bayesian methods for decision making.
The content is applicable to any process, but focus will be on applications in the pharmaceutical industry with examples from all three stages of the Process Lifecycle: Design, Qualification, and Ongoing Monitoring.
21st-Century Design of Screening Experiments
Peter Goos, KU Leuven, Belgium
Many industrial experiments are screening experiments. In the past few decades, much innovative work has been done on screening experiments involving two levels per factor as well as screening experiments involving three levels per factor. As a result of all that research, many attractive two-level orthogonal arrays are now available as alternatives to the traditional two-level fractional factorial designs, and definitive screening designs, involving three levels per factor, have become available as well.
In this course, these new screening designs will be discussed in detail, and compared to the well-known classical alternatives. Attention will be paid to the existence of non-isomorphic orthogonal arrays and definitive screening designs. Also, the various criteria that can be used to evaluate and to compare the modern screening designs, such as generalized resolution and generalized aberration, will be introduced in an intuitive, graphical way. Therefore, a major goal of the course is to provide an accessible overview of the recent developments concerning nonregular orthogonal screening experiments. Finally, the course will also discuss the usefulness of optimal screening designs as alternatives to orthogonal screening designs.
Strategies for Formulations Development
Ronald D. Snee, Snee Associates LLC
Roger W. Hoerl, Department of Mathematics, Union College
This course will help scientists and engineers develop formulation recipes more quickly and efficiently. Good strategies are needed to get the right data in the right amount at the right time. Ron Snee and Roger Hoerl have worked on this problem for decades and have executed many different types of formulation and mixture experiments in a variety of industries. This experience has enabled them to identify what is essential for successful experimental designs, rather than what is just nice to know. Participants will be able to immediately apply what they have learned.
Key takeaways from the course include learning how to:
- Approach formulation development from a strategic viewpoint, where the resulting experiments provide a roadmap for developing a successful formulation.
- Focus on developing understanding about how components blend together.
- Use designs and models that focus on finding dominant components with large effects.
- Use screening experiments to identify components that are most critical for formulation performance. This strategy ensures that important components are not overlooked.
- Analyze both screening and optimization experiments using graphical and numerical methods. The right graphics can extract additional information from the data.
- Integrate formulation components with process variables in designs and models, using recently developed methods that reduce the required experimentation by up to 50%.
How to reduce mistakes, better meet deadlines, avoid wasted experimentation and reduce the cost of experimentation in terms of time, personnel and funds.
Bridging Statistics and Data Science
Ming Li, Amazon
With the recent big data revolution, enterprises ranging from FORTUNE 500 to startups across the US are using Data Science to bring valuable business insight from all the data collected. Statisticians are great data scientist candidates, but there are relatively few data scientists with statistics education background. This course aims to bridge the gap between Statistics and Data Science. Data science is a combination of science and art with data as the foundation. We will cover both the science part and the art part (such as data science project flow, general pitfalls in data science projects, and soft skills to communicate with business stakeholders effectively). The course will be hands-on and we will use the Databricks community edition cloud platform and R-Studio to illustrate programming, big data platform usage (such as Spark) and standard machine learning algorithms.