|
|
IIn
this project we explore novel technologies for efficient
summarization and sensemaking based on dynamic data from complex
processes. This research is motivated by emerging advanced
infrastructures that facilitate rapid operational data
collection (e.g., bedside medical devices, energy monitoring
hardware, data acquisition products based on wireless sensor
networks, etc.). This project includes the following tasks:
Task 1: Process Data
Warehousing.
Task 2: Signature-based Analysis
of Numeric Process Logs.
Task 3: Assessment of Process Dynamics based on
Information Divergence.
Our patented Process Data Warehouse (PDW) technology deals with
the task of efficient utilization of large-scale numeric data
streams. The PDW implements a novel approach to continuous
summarization and discovery of trends in dynamic process data.
The proposed approach includes a method, system architecture and
a set of optimization techniques for efficient warehousing of
large data streams. Potential uses of the PDW technology range
from data utilization in specialized data centers to
Internet-scale data search and analysis engines. A PDW system
can explore the most and least probable scenarios in development
of complex processes that produce large amounts of raw data. The
data summaries generated by the PDW system can further be used
for intelligent assistance with the process monitoring and early
warning. For example, taking into account summarized information
on energy consumption the system would be able to generate early
warnings of high probability of a power outage. Similarly it can
be used in many other application domains (e.g., medical data
analysis, market data monitoring, natural disasters and
structural health analysis, etc.).
A major challenge in large-scale process monitoring is to
recognize significant transitions in the process conditions and
to distinguish them from random fluctuations that do not produce
a notable change in the process dynamics. Such transitions
should be recognized at the early stages of their development
using a minimal ``snapshot'' of the observable process log. We
consider a novel approach to detecting notable transitions based
on analysis of coherent behavior of frequency components in the
process log (coherency portraits). We have found that notable
transitions in the process dynamics are characterized by unique
coherency portraits, which are also invariant with respect to
the random process fluctuations. Our experimental study
demonstrates the significant efficiency of our approach as
compared to traditional change detection techniques.
Process monitoring involves tracking system’s behavior,
evaluating the current state of the system, and assess its
severity. In this task we develop an approach based on
steady-state analysis of the process model generated from a
numeric process log. In particular, we utilize Markov process
models to continuously produce steady-state vectors reflecting
the process dynamics at different time moments. We explore
measures of information divergence to assess deviations in a
sequence of steady state vectors as process evolves. This
approach is efficient in detecting and early warnings about
severe deviations in the expected process behavior.
PhD
Students:
Andrii Cherniak
Yihuang Kang
Ying-Feng Hsu
Selected References
-
Cherniak,
A., Zadorozhny, V. Signature-Based Detection of Notable
Transitions in Numeric Data Stream. To appear in IEEE
Transactions on Knowledge and Data Engineering, 2013.
-
Kang, Y.,
Zadorozhny, V. Divergence-based Detection of Severe Process
States. Under preparation.
-
Zadorozhny,V. Process Data Warehouse. US
Patent 7,933, 861 issued on April 26, 2011
-
Bickel,
J., Visweswaran, S., Levin, J., Hsu, Y-F.., Kang, Y.,
Zadorozhny, V.
. Data
Warehousing and Markov Modeling of Children Admitted with
Respiratory Complaints. American Medical Informatics
Association Symposium, San Francisco, 2010
-
Hsu, Y.-F. Efficient Information Processing Architecture for
Early Warning Systems. Dissertation Proposal, Graduate
Information Science and Technology Program, School of
Information Sciences, University of Pittsburgh, 2010
Complete
list of publications
|
![]() |
DIP: Data-Intensive Process Monitoring
School of Information Sciences 
University of Pittsburgh