DMIP 2017

Workshop Description

All fields where data is collected have seen an increase in the amount of data being analyzed, and this has led to an increase in work related to data mining and machine learning. These methods are moving out of academic and high tech fields and into new and everyday applications. This half day workshop held in conjunction with the 2017 IEEE International Conference on Data Mining (ICDM) will focus on two primary issues when applying data mining in practice: how to incorporate the cost of data into the problem and how to automate aspects of the data mining process.

Cost

From a data perspective, cost can be divided into three primary areas: 1) collection cost – costs associated with acquiring each data stream, 2) labeling costs – costs associated with assigning classes, acquiring response variable values in supervised learning, or acquiring the values of missing explanatory data, and 3) processing costs – costs associated with model training, prediction, and storage. In the data mining literature, algorithms and models are usually optimized with respect to predictive accuracy and little is published on incorporating costs into the data mining process. However, in almost all real-world data mining applications, costs are present and should be considered. In order to data mining applications to be successful, they must meet the system requirements and objectives while not exceeding the budget.

Automation

There have been many debates in recent years about the need and the ability to automate data mining and machine learning tasks. A recent blog post titled “Data Scientists Need More Automation” discusses the repeated efforts required to configure and run services or scripts on a network of machines. Other discussions ask, “Can We Automate Data Mining?,” arguing that many tasks performed by data scientists require manual intervention and thus cannot be automated; in other words, expertise is needed for each individual case, requiring clear understanding of the business and the data. The development of tools to automate data mining efforts fosters the transformation of theory to application and also promotes the development of standards and the adoption of these standards. Automated standards enable researchers and practitioners to better communicate, sharing successes and challenges in a more consistent common language. In an age of software as a service and ever-increasing scalability requirements, standards are necessary. Consistent adoption, application, and communication in turn promote research and refinement of the automated strategies and growth of the community. The challenges that must be discussed relate to the boundaries of automated tasks and individual attention needed for each unique business and data scenario.

Topics include (but are not limited to):

Automation • Automated methods in machine learning, data mining, predictive analytics, and deep learning • Automated methods in knowledge discovery in databases • Automation theory, automation and optimization • Hyperparameter autotuning • Automated pipelines and process-flows in production systems • Automated approaches to model monitoring and updating • Automated methods for streaming data • Internet of Things (IoT) and automation • Automated data preparation, automated variable and model selection • Automation in big data, automation in massive modeling

Cost • Active learning and cost • Missing data algorithms • Feature selection and cost • Efficient feature engineering • Algorithm processing time analysis • Model training algorithms that incorporate costs • Data mining on a budget applications in real-world systems and environments • Theory of data mining on a budget • Metrics that combine accuracy and cost

Submission Instructions

Paper submissions should be limited to a maximum of eight (8) pages, in the IEEE 2-column format, including the bibliography and any possible appendices. All papers must be formatted according to the IEEE Computer Society proceedings manuscript style, following IEEE ICDM 2017 submission guidelines. Microsoft Word templates and LaTeX templates are available at the following web page:

http://www.ieee.org/conferences_events/conferences/publishing/templates.html

We do not accept email submissions. Authors are invited to submit the papers electronically in PDF format through the IEEE ICDM CyberChair online submission system:

• Automation papers: https://wi-lab.com/cyberchair/2017/icdm17/scripts/submit.php?subarea=SP29 • Cost papers: https://wi-lab.com/cyberchair/2017/icdm17/scripts/submit.php?subarea=SP31

All submissions will be double-blind reviewed by the Program Committee following IEEE ICDM 2017 guidelines. Author names and affiliations must not appear in the submissions.

Organizing Committee

Automation Tao Wang, SAS Institute Inc., t.wang@sas.com Patrick Koch, SAS Institute Inc., patrick.koch@sas.com Brett Wujek, SAS Institute Inc., brett.wujek@sas.com

Cost Stephen Adams, University of Virginia, sca2c@eservices.virginia.edu Peter Beling, University of Virginia, beling@virginia.edu William Scherer, University of Virginia, wts@eservices.virginia.edu