Data mining: methods and applications

1
Data mining methods and applications

• Heikki Mannila
• FDK Scientific Advisory Board Meeting Nov. 28,
  2002

2
Why?

• Lots of data
• Novel types of data
• Computational techniques gaining in importance
• Algorithmic and probabilistic methods are needed

3
Approach and goals

• Application area ? concept formation ?
  algorithmic question ? algorithm ? analysis ?
  back to practice
• Algorithmic basic research in areas where the
  results can be put into practice
• Understanding the fundamental properties of data
  summarization and analysis

4
Computational methods

• Pattern discovery
• Finding recurrent patterns in large data sets
• Succinct representations of data
• Combinatorial algorithms
• Dynamic programming etc.
• Probabilistic modeling and analysis (Markov Chain
  Monte Carlo)
• Combining pattern discovery and combinatorial
  algorithms

5
Application areas

• Genome structure
• Understanding the types of variation in genomes
  (within and between species)
• Gene mapping
• Gene expression
• Document data
• Telecommunications data
• Spatial data (onomastics)

6
Research topics methods

• Finding structure in large 0-1 data sets
• Similarity between complex data objects and
  descriptions
• Foundations of data analysis
• Condensed representations for classes of queries
• Probabilistic logic and its uses
• Pattern discovery and Bayesian analysis
• Combining different types of data (sequence,
  spatial, temporal)

7
Structure in 0-1 data sets

• I.e., document data
• Find collections of words that tend to co-occur
  topics / m-out-of-n concepts
• Patterns and probabilistic modeling
• Leads to questions in generalizing Bonferronis
  inequality
• Distances between representations

8
Research topics applications

• Genome structure
• Haplotype blocks how is the variation in the
  human genome structured?
• Finding recurrent sources in sequences

9
Haplotype blocks

• Is the variation between the genomes of
  individuals distributed uniformly?
• Data aligned 0-1 sequences
• Task find whether the data have a block
  structure
• Concept block structure segmentation centers
  variation

10
Blocks, cont.

• Quality of structure description length
• Algorithm Dynamic programming and clustering ?
  block boundaries
• Dynamic programming ? estimates for the strength
  of block boundaries

11
(No Transcript)
12
Finding recurrent sources for sequences

• Data long sequences (genome, telecom,
  paleoecology)
• Problem do some parts of the sequence come from
  the same source?
• Probabilistic model underneath
• Concept (k,h)-segmentation

13
(k,h)-segmentations

• A (multidimensional) time series
• A segmentation of the sequence into k pieces, but
  with at most h different levels
• Normal segmentation into homogenous pieces
  (k,k)-segmentation

14
(k,h)-segmentation, cont.

• NP-hard in the general case
• Status unknown for dimension 1
• -approximation algorithm for L2
• 3-approximation for L1
• Simple and practical algorithms
• Good performance
• Results in applications

15
(No Transcript)
16
Future topics

• Theory and applications of data mining
• Combinatorial algorithms with probabilistic
  notions statistics
• Foundations condensed representations, machine
  learning and probabilistic logic
• Spatial and spatio-temporal data
• Genome structure
• Linguistic data, process data,

17
(No Transcript)