Secure Knowledge Management

1
Secure Knowledge Management
Shouhuai Xu (UTSA) Weining Zhang (UTSA) Ravi
Sandhu (GMU)

• Motivation DataBase Management Systems (DBMS)
  have been widely and successfully used to manage
  enterprises data assets. Due to privacy concerns
  and regulations, data cannot be shared among
  parties as one may have wanted. This means that
  the knowledge (e.g., decision trees) hidden in
  the data cannot be shared. This project envisions
  an industry in which knowledge service providers
  may extract knowledge from many owners data to
  serve customers need for knowledge in their
  business activities.

Plot on right is an architectural framework for
secure knowledge management systems. The datasets
are owned by parties that do not trust each
other. The knowledge extractor may be run by
parties who do not have any data. The extracted
knowledge is served via the knowledge servers,
which disseminate knowledge to the knowledge
consumers. Any appropriate business models can be
incorporated into this framework.
Graphics goes in this box. Graph captions in
Helvetica, 36pt, white.
Approach and Impact

• Research Impact
• Models, architectures and mechanisms that can be
  used to found secure knowledge management systems
  and could be used for other applications

• New Approaches
• Merge-based and incremental and cryptographic
  knowledge extractions
• Data-private, breaching- and scalability-aware
  knowledge dissemination

• Research challenges on extraction
• Analyze incremental knowledge extraction (ongoing
  preliminary results seem positive).
• Analyze merge-based knowledge extraction (ongoing
  preliminary results seem positive).
• Design and analyze new perturbation methods for
  secure knowledge extraction.
• Design practical and provably-secure
  cryptographic protocols for knowledge extraction
  involving a large number of large datasets.

• Research challenges on dissemination
• (Impossibility result, informally stated) Any
  real-life knowledge models (e.g., decision trees)
  are subject to knowledge breaching attack.
• How can we develop models to quantify knowledge
  breaching? In what metrics? Breaching bounds?
• How can we design mechanisms to match the
  knowledge breaching bounds?
• Design practical cryptographic protocols to
  fulfill data-private, policy-driven,
  scalability-aware knowledge dissemination.