A guide to AI, machine learning and new workflow technologies at HIMSS17 Part 1: Machine learning and workflow

Charles Webster, MD, (a HIMSS17 Social Media Ambassador and @wareFLO on Twitter) comprehensively researched which vendor is showing what and where. Part one in his three-part series focuses on the workflow technology increasingly powering data science.
By Charles Webster, MD
07:11 AM
Share
AI machine learning workflow HIMSS17

I have been tracking diffusion of workflow technology into healthcare for over two decades. Since 2011 I’ve annually searched every HIMSS conference exhibitor website for workflow-related material. I’ve seen emphasis on workflow go from very little to a lot, and mentions of workflow engines and business process management (BPM) go from almost zero to substantial. This kind of workflow technology has been around for decades outside of healthcare. It is finally flowing into healthcare and health IT in a big way.

Further, more healthcare organizations are beginning to model their workflows using BPMN (Business Process Model and Notation), evidenced by a recent workshop on healthcare and BPMN I attended. Since many BPM suites execute BPMN, the next step will be to use BPMN to build modern healthcare workflow systems. This is the kind of healthcare workflow technology about which I usually write, tweet, and present.  It is important and will become even more important in healthcare.

However, as Wil van der Aalst, a leading BPM researcher pointed out to me in a 2013 interview, workflow management systems and business process management suites are not the only examples of what academics call process-aware information systems. In this series I will discuss process-awareness in data science and machine learning, conversational user interfaces and chatbots, and containers and microservices, all of which are increasingly represented among HIMSS conference exhibitors.

Business process management exhibitors at HIMSS17 include Alfresco (booth 7853), Appian (4376), and Fujitsu (351).

Data science and machine learning
HIMSS17 exhibitor profiles mentioning data are prolific (545, at the time I write this). Only seven specifically mention “machine learning,” but you can bet most vendors with access to lots of healthcare data are beginning to experiment with ML, especially since Google released the open source TensorFlow deep learning software.  I downloaded and had TensorFlow up and running on test data in just a few minutes. I should mention I used the very first textbook on Machine Learning (An Artificial Intelligence Approach) during graduate school, where its editor Ryszard Michalski taught. I also took courses on neural networks with some of the graduate students who laid the foundations for deep learning (hidden layer and recurrent neural networks).

Machine learning, training computer programs to improve performance, is part of data science. Both data science and machine learning increasingly use workflow technology to effectively and efficiently turn data into action. Terminology varies; sometimes data workflow models are called data flow or pipeline models. Nevertheless, what we are talking about here is process-awareness in the sense workflow and BPM researchers have been studying for decades. There is a model, or representation, of a workflow, process, or data flow. And there is a software platform or engine interpreting the model to facilitate data science and machine learning goals.

For example, the following is a typical data science workflow: load data, understand data, create data objects, train model, make predictions, and compute error. It is not important for you to be a data scientist to understand this workflow, merely to understand this it is indeed a workflow, a sequence of activities (loading, training, calculating), consuming resources (data, time, attention), and achieving goals (finding useful patterns in data).

Machine learning also relies on workflows. I won’t delineate them here. But they resemble data science workflows in the following ways. In both cases, workflows are specific tasks sequentially linked together and automated. Both data science and machine learning involve repetitive combinations of software-based and human manual-based operations. By automating data science and machine learning tasks, by extracting variables from the workflows and managing those variables using workflow, data flow, or pipeline management technology, results become more consistent, accurate, valuable, reusable, and shareable. There is even a new machine learning workflow language, called WhizzML, to better achieve these important goals of automating machine learning workflows.

Companies mentioning machine learning in their HIMSS17 profiles include CareSkore (Booth 1623), DB Networks (376-36), Edgility (7785-12), Invincea (7785-13), OM1 (7844), Prophit Insight (7785-88), and QPID Health/Evicore Healthcare (5574, VHQ4400). Their uses of machine learning range from identifying at-risk patients to detecting network security threats.

Tomorrow, in my second installment, I will show how conversation “workflows” makes chatbots possible. 

HIMSS17 runs from Feb. 19-23, 2017 at the Orange County Convention Center.


This article is part of our ongoing coverage of HIMSS17. Visit Destination HIMSS17 for previews, reporting live from the show floor and after the conference.


Like Healthcare IT News on Facebook and LinkedIn