haishan liu 1, gwen frishkoff 2, robert frank 1, dejing dou 1 1 university of oregon 2 georgia state...
Post on 19-Dec-2015
217 views
TRANSCRIPT
- Slide 1
- Slide 2
- Haishan Liu 1, Gwen Frishkoff 2, Robert Frank 1, Dejing Dou 1 1 University of Oregon 2 Georgia State University
- Slide 3
- ERP (Event-Related Potentials): a direct measure of neuronal activity Lack of meta-analysis across experiment NEMO (Neural ElectroMagnetic Ontologies) for data sharing and integration Goal of the presented study Mapping alternative sets of ERP spatial and temporal measures
- Slide 4
- Alternative sets of ERP metrics
- Slide 5
- Semi-structured data
- Slide 6
- Uninformative column headers
- Slide 7
- Semi-structured data Uninformative column headers Numerical values
- Slide 8
- Cluster labels Meaningful labels Point-sequence curve
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Process all point- sequence curves Calculate Euclidean distance between sequences in the Cartesian product set (Cross-spatial join) Metric Set1 Metric Set2
- Slide 13
- Slide 14
- the two datasets contain the same or similar ERP patterns
- Slide 15
- 1-to-1 mapping between metrics
- Slide 16
- the two datasets contain the same or similar ERP patterns 1-to-1 mapping between metrics Minimum sum of distances 4.01 + 3.74 > 4.08 + 3.57
- Slide 17
- Slide 18
- Wrong Mappings. Precision = 9/13 Gold standard mapping falls along the diagonal cells
- Slide 19
- 3-Factor design of experiment data (Fully factorial: 2 x 2 x 2) 2 simulated subject groups (samples) SG1 = sample 1 SG2 = sample 2 2 data decompositions tPCA = temporal PCA decomposition sICA = spatial ICA decomposition 2 sets of alternative metrics m1 = metric set 1 m2 = metric set 2
- Slide 20
- Overall Precision: 84.6%
- Slide 21
- Use of an ontology to assign meaningful labels to ERP patterns Application of sequence similarity search in discovering mappings across alternative metrics Extension of the instance-level approach in schema matching Articulation of a global minimum heuristic
- Slide 22
- Questions and comments? Please contact Haishan Liu ([email protected])