Some Interesting Problems

1
Some Interesting Problems

• Rakesh Agrawal
• IBM Almaden Research Center

2
Long Sequential Patterns

• Bottoms-up algorithms such as Apriori (and its
  variants) blow up
• For potential approaches, see
• Bayardo Sigmod 98
• Bayardo, Agrawal KDD 99
• Han, Pei Talk in this workshop

3
Actionable Patterns

• Principled use of domain knowledge
• Discard uninteresting patterns
• Performance
• See papers in the recent KDD conferences for
  ideas

4
On-line Algorithms

• Can touch a data item k times in a time window w
• Incremental update of the model
• When to retire the data?

5
Richer Patterns

• Discovering trees and dags
• Medical Research
• Heikki et al Episodes

6
Mining over privacy-preserved data

• Randomization
• Replace x by xy where y is drawn from a a known
  distribution
• With probability p, replace a by some other
  categorical value not present in the transaction
• Anonymization

7
Asking questions easy