A series of webinars the 3rd week of each month, Sep-Nov, 10:00am - 12:00 pm EDT (4 - 6:00 pm CET)
November 19, 2020: Statistical Engineering: What It Is, What It Is Not, and How it Relates to Other Disciplines – A Member’s Perspective
In this session, invited speakers and panelists provide their own perspectives on the emerging field of statistical engineering and describe how they believe it relates to and differentiates from other disciplines. Through this open venue, we hope to sharpen how we articulate our motivation for establishing ISEA as a unique society to promote the development of the field of statistical engineering
October 22, 2020: Statistical Engineering Case Studies: University Experiences.
After setting the stage by discussing the importance of case studies to the discipline of Statistical Engineering, we then discuss two University projects. The first application comes from the University of Amsterdam’s consultancy group with a case study on how to predict student success or failure using machine learning and applying statistical engineering theory. The second case comes from a Statistical Engineering course at Virginia Tech, where the students gained real-world experience dealing with complex and unstructured problems dealing with Covid-19. (Recording available by clicking the title of the presentation.).
Martha Gardner (GE): "Case Studies for Statistical Engineering"
Case studies have long been useful tools in helping people new to a discipline understand the unique aspects of the discipline and how to best utilize the approaches of the discipline in their daily work. In this talk, I will discuss the importance of case studies in understanding the new discipline of Statistical Engineering and how the International Statistical Engineering Association is compiling and sharing these case studies with the broader community.
Leo C.E. Huberts and Ronald J.M.M. Does (Department of Operations Management, University of Amsterdam, the Netherlands): "Statistical Engineering and Machine Learning: A Case Study to Predict Student Success or Failure"
A quote from a high school principal: “Early Warning Indicator Reports were invaluable to the success of our School”. These reports monitor students throughout their school career and warn teachers and staff of students with high dropout risks. Also, monitoring allows for the identification of students who are insufficiently challenged and will benefit from more stimulating classroom material. Several approaches for monitoring student progress are evaluated to answer three research questions related to this problem:
(1) What determines student performance?(2) How can statistics and machine learning tools be used in monitoring student progress?(3) Which methods can be used for predictive monitoring of student results?
The authors will share how they worked together with a Dutch high school and combined hierarchical Bayesian modeling with statistical and predictive monitoring procedures. And, how machine learning tools (recurrent neural networks) were applied . The final results give a clear blueprint for student progress monitoring as documented in this report.
Geoff Vining (Virginia Tech):"Teaching Students How to Address Complex, Unstructured Problems: Covid-19 Project with Socially Determined"
The future of industrial and corporate statisticians lies in finding workable solutions to very complex and unstructured problems. Unfortunately, the standard university curriculum fails to address this very issue. In fact, the standard curriculum gives the impression that all problems are well-structured and that mathematically optimum solutions always exist. As a result, students often have trouble dealing with complex problems and finding reasonable, workable, but not optimal in any mathematical sense solutions. This talk summarizes one instructor’s attempt to give graduate students in statistics some grounding in such problems. The capstone experience was working with Socially Determined early in the pandemic to advise the governors of Virginia and Maryland on how to spend Federal money to relieve the impact of the virus on their economies. The key was to determine how to use the money most effectively. The students made a significant contribution to the overall project.
September 17, 2020: Bringing Statistical Engineering to Life via Case Studies
Discover how Statistical Engineering is applied via three real-world case studies. These presentations highlight how this emerging methodology is already being used to improve the world around us. (Recording available by clicking the title of the presentation.)
The expected cost of quality assessment delay in large scale industrial units is high. Expected consequences include an increased risk of producing large amounts of out-of-spec material, poor process control (process instability and high product variability), increased logistical costs, delayed product release, among others. Soft sensors or inferential models offer a possible path to shorten the product quality assessment cycle, opening windows of opportunity to mitigate these effects. Different types of soft sensors can be conceived depending on the goal (causal models or correlation-based), available data (process data, measurements from Process Analytical Technology devices, images, etc.), type of processes (continuous, batch). These and other aspects related to soft sensor development and use will be discussed in this talk, where several applications in different sectors are also referred (petrochemical, pharmaceutical, semiconductors).
Computer vision is often used in inspection processes to discriminate between good and bad product, or to separate many fractions that constitute a material stream (e.g. in food or plastics inspection). In order to build a statistical model that can be used to perform the inspection task, typically a mixture of fractions is presented to the camera system and a human operator then labels the different fractions in the acquired images. This manual, on-screen labeling of images is prone to errors, so that wrongly labelled data (what we call label noise) corrupt the obtained database to build a model, potentially leading to inferior classification results. Data can also be corrupted by measurement noise – measurements that are noisy or non-representative.
In this talk, we will discuss how both sources of noise affect the performance of industrial classifiers. Furthermore, as a potential solution, we introduce a fast yet robust algorithm that is capable of withstanding large amounts of noise. A practical example is used to demonstrate the approach we propose.
The Procter & Gamble Company (P&G) is one of the top 10 largest consumer packaged goods companies and is considered one of the leading companies contributing to growth and innovation in an evolving market. In order to stay competitive, it is essential for P&G to play at the leading edge of product superiority by providing consumers with high-performing options. This involves not only leading the market on key benefit spaces but also communicating those benefits to consumers around the world. P&G has the NA Competitive Product Laundry Project across the myriad of consumer benefit spaces in Laundry Category: Stain Removal, Odor Removal, Whitening, and other attributes. The initiative meets the criteria of a large, complex, unstructured problem laid out in Hoerl and Snee (2017) that would benefit from the strategies of Statistical Engineering. In this talk, we discuss how each of the elements of Statistical Engineering, 1) Identify the high impact problems, 2) Providing structure, 3) Understanding context, 4) Develop Strategy, 5) Develop and execute tactics, and 6) Identify and deploy a final solution, were leveraged in the success of the initiative.