Over the past decade, the field of machine learning has made great strides in developing computational methods for extracting predictive relationships from data. There have been many advances in supervised learning from labelled training cases, and unsupervised discovery of regularities in data. These areas have seen steady progress on both the theoretical and practical fronts, with many learning techniques now being used regularly in commerce and industry.
Unfortunately, this emphasis on prediction has meant little effort has been devoted to other important aspects of data analysis. In particular, there has been relatively little work on methods for discovering anomalies or outliers in data, which are irregularities that cannot be explained by existing domain models or knowledge. Furthermore, much of the existing work focuses on detecting outliers solely for the purpose of removing them from the analysis to prevent them from unduly affecting learning instead of treating them as interesting phenomena in their own right.
Despite the community's focus on other areas, some researchers have designed, implemented, and evaluated systems that use machine learning techniques for discovering anomalies. However, they are spread across a number of communities and publish in different venues. These include fields like patient monitoring, detecting disease outbreaks, spacecraft monitoring, computer intrusion detection, geographic information systems, and accounting fraud detection. Because of this topic's importance, it seems important to bring together these distinct groups to exchange ideas and lessons.
To this end, we have organized a symposium on machine learning
for anomaly detection. The meeting will bring together
researchers from these different areas, letting them report their
recent results and discuss common concerns. This interaction should
pave the way for increased work on this topic and collaborations
between some of the active groups.
The symposium will take place on Saturday, May 22, and Sunday, May 23 at Stanford University's Center for the Study and Language and Information (CSLI). Talks will be held in the main conference room of Cordura Hall on the Stanford campus.
There is no registration fee for the symposium, but attendance will be by invitation only. There will be 10 invited speakers presenting at the meeting over two days. We will have space for some non-presenting attendees at the meeting. If you are interested in participating, please send email to langley@csli.stanford.edu with a brief account of your previous and current work on the symposium topic.