Closing Remarks

1
Closing Remarks

• DUNE 2007
• Reynold Cheng

2
Probabilistic Databases

• Managing uncertainty in real-world applications
  is important
• Capture correlations with graphical models
• for different granularities
• Storing attribute uncertainty as tuple
  uncertainty
• Efficient data integration (through incremental
  computation of Bayesian networks)

3
Uncertainty in Bioimage Databases (Keynote by
Prof. Ambuj Singh)

• Different level of uncertainty
• pixel, shape
• Represented by APLA
• Queries
• Range query, NN query, joins
• Mining
• Segmentation and convex hull generation
• What is the suitable level of abstraction for
  computer scientists?

4
Pattern Discovery

• Reducing UK-means to K-means
• for mean squared error distance metric
• UF-tree for mining uncertainty in transactions

5
Bayesian networks classification

• Neural network that handle geostatistics for
  spatial applications
• Dynamic Bayesian networks
• Segment multi-attribute sequences
• Future models should deal with missing data
• Acoustic bird species recognition
• Skewed class distributions and mislabeled example
  evaluation

6
Discussions (1)

• Development trend in database community
• probabilistic databases
• focus on query and system performance
• Data mining
• Techniques for data cleaning and modeling
• Can we combine the merits of these 2 areas?
• Adopt appropriate uncertainty models and develop
  high-performance DB systems

7
Discussions (2)

• The theories and models for probabilistic
  databases are more mature
• However, there lacks a killer application that
  maximizes the potential of probabilistic DBs
• Time to think about how uncertainty management
  can be used by applications
• e.g., how can they be used in bioimage DBs?