Strategic Insight from Machine Learning
Digitalization creates high speed, high volume digital activity traces – Big Data. Using recent advances in machine learning, organizations can use these data to automatically build models that predict specific events in real-time. Managers can use machine learning-based analytics to improve a wide variety of tasks ranging from targeted marketing to preventative maintenance to fraud detection. The course emphasizes processes for approaching and defining problems assessable through machine learning. The class will function like a lab in which students get hands-on experience with cutting-edge analytics processes, tools, and techniques. This firsthand experience provides students with insight into how big data analytics can improve business processes and decision-making.
This is an intensive, 4-day course. Before the course begins, students must read a background text to ensure a common vocabulary and familiarity with basic techniques. Most of the course will occur like a workshop, in which we collectively explore tools and reflect on their strengths, weaknesses, and potential strategic application. Because discovery/surprise are so important in this process, most slides and materials will NOT be made available in advance of class.
Day 1: Defining ML and Isolating ML-suitable problems
We will spend day 1 specifying what machine learning (ML) is, what it can do, and how it interfaces with the larger organization. We will answer questions including: What is machine learning (ML)? How is ML distinct from classic scientific inquiry? From other elements of the cognitive computing universe? What are the fundamental techniques/algorithms in ML? With this definition of ML and its techniques, to what types of problems does the toolset lend itself? How does ML interface with the organization’s larger digital ecosystem?
Day 2: Data, Modeling, Model Assessment, and Automated Modeling
This day will present the “nitty gritty” of how machine learning is done – we will crack open the hood, open up the engine, and see how the tool works. We will begin the day reviewing data storage, collection, and transformation. We will then execute some modeling exercises stepwise before a discussion about what constitutes a “good” model. This discussion will cover important technical topics such as overfit, cross-validation, and target leak. Finally, we will introduce automated ML.
Day 3: More on Auto-ML
In day 3, we will introduce tools to conduct machine learning, which we will use to conduct several illustrative machine learning projects. Distinguishing model evaluation from model assessment, we will introduce techniques to assign value to models, and discuss that as a bridge between granular model output and higher-level strategic insight.
Day 4: Sources of Resistance, Other Tools, Other Problems, and Next Steps
On day 4, we will discuss auto-ML tools and techniques other than auto-ML that lend themselves to different types of problems (i.e., other spaces in the cognitive computing universe). We will also discuss sources of cultural and institutional resistance to ML adoption, and strategies to handle such resistance. We will have a final student exercise, in which students apply their learning and present strategic insights from a machine learning project. Finally, we will discuss next steps for ourselves and for industry generally.
This is an excerpt from the complete course description for the course. If you are an active student at BI, you can find the complete course descriptions with information on eg. learning goals, learning process, curriculum and exam at portal.bi.no. We reserve the right to make changes to this description.