Learning Spatial Decision Trees for Geographical Classification Student: Zhe Jiang Advisor: Prof. Shashi Shekhar Thesis Committee Members: Prof. Shashi Shekhar, Prof. Vipin Kumar, Prof. Arindam Banerjee, Prof. Joseph Knight, Prof. Snigdhansu Chatterjee 1
31
Embed
Learning Spatial Decision Trees for Geographical Classification Student: Zhe Jiang Advisor: Prof. Shashi Shekhar Thesis Committee Members: Prof. Shashi.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Learning Spatial Decision Trees for Geographical Classification
Student: Zhe JiangAdvisor: Prof. Shashi Shekhar
Thesis Committee Members: Prof. Shashi Shekhar, Prof. Vipin Kumar,
Prof. Arindam Banerjee, Prof. Joseph Knight,Prof. Snigdhansu Chatterjee
2
Biography
• Education:– PhD student in Computer Science, University of Minnesota, 2010 – now– B.E. in Electrical Engineering, University of Science and Technology of
China (USTC), 2006 – 2010
• Current Project:– Understanding Climate Change: A Data Driven Approach (2010 - now )
• Awards:– Doctoral Dissertation Fellowship, University of Minnesota, 2015 – 2016– NSF Travel Awards for SSTD 2011, ACM GIS 2012, IEEE ICDM 2014
3
Thesis related publications:[1] Jiang, Zhe, Shashi Shekhar, Xun Zhou, Joseph Knight, and Jennifer Corcoran. "Focal-Test-Based Spatial Decision Tree Learning." IEEE Transactions on Knowledge & Data Engineering (TKDE) 6 (2015): 1547-1559.
[2] Jiang, Zhe, Shashi Shekhar, Xun Zhou, Joseph Knight, and Jennifer Corcoran. "Focal-test-based spatial decision tree learning: A summary of results." In Data Mining (ICDM), 2013 IEEE 13th International Conference on, pp. 320-329. IEEE, 2013.
[3] Jiang, Zhe, Shashi Shekhar, Azamat Kamzin, and Joseph Knight. "Learning a Spatial Ensemble of Classifiers for Raster Classification: A Summary of Results." In Data Mining Workshop (ICDMW), 2014 IEEE International Conference on, pp. 15-18. IEEE, 2014.
[4] Jiang, Zhe, Shashi Shekhar, Pradeep Mohan, Joseph Knight, and Jennifer Corcoran. "Learning spatial decision tree for geographical classification: a summary of results." In Proceedings of the 20th International Conference on Advances in Geographic Information Systems (GIS), pp. 390-393. ACM, 2012.
Other selected publications:[5] Shekhar, Shashi, Zhe Jiang, Reem Ali, Emre Eftelioglu, Xun Tang, Viswanath Gunturi, Xun Zhou, “Spatiotemporal Data Science: A Computational Perspective”, Special Issue on Advances in Spatio-Temporal Data Analysis and Mining, ISPRS International Journal of Geo-Information, 2015. (minor revision)
[6] Ramnath, Sarnath, Zhe Jiang, Hsuan-Heng Wu, Venkata MV Gunturi, and Shashi Shekhar. ”A Spatio-Temporally Opportunistic Approach to Best-Start-Time Lagrangian Shortest Path.” In Advances in Spatial and Temporal Databases (SSTD), pp. 274-291. Springer, 2015.
[7] Mohan, Pradeep, Shashi Shekhar, James A. Shine, James P. Rogers, Zhe Jiang, and Nicole Wayant. “A neighborhood graph based approach to regional co-location pattern discovery: A summary of results.” 19th ACM SIGSPATIAL GIS, pp. 122-132. ACM, 2011.
[8] Jiang, Zhe, Micheal Evans, Dev Oliver, Shashi Shekhar. “Identifying K Primary Corridors from Urban Bicycle GPS Trajectories on a Road Network.” Special Issues on Mining Urban Data, Information Systems Journal, Elsevier (major revision).
WPE exam
this talk
Publications
4
Outline
• Motivation• Problem Statement• Challenges• Related Work• Proposed Approach• Evaluation• Conclusion, Future Work
5
Motivation• Civil earth observation – a national priority
– Geo-referenced digital information about Earth
– Societal benefit areas
• Other potential application– lesion classification and brain tissue segmentation in MRI images
• High computational cost– large amount of computation with spatial
neighborhoods of different sizes
Decision tree predictionGround truth classes
13
Related Work & Limitations
Decision tree classifiers
Traditional non-spatial tree
Spatial entropy and information gain
Local-test-based decision tree Focal-test-based decision tree
(ID3 1986, CART 1984, C4.5 1993)
(Jiang 2012, Li 2006, Stojanova 2011 & 2012)
o i.i.d. assumptiono ignoring spatial autocorrelationo salt-and-pepper noise
o tree nodes test each pixel independentlyo spatial autocorrelation in selection of node testo still salt-and-pepper noise when all candidate tests poor
True class map Local test (numbers), fixed neighborhood (blue)
Local test (numbers),adaptive neighborhood(blue)
Focal test results and predictions
Focal test results and predictions
19
Proposed Approach: Computational Refinement
• Computational bottleneck: number of focal computation– quadratic to number of samples– linear to number of distinct feature values– linear to number of features
• Key idea– sort all candidate test thresholds (feature values) in order– focal function values mostly same across two thresholds– Cross-threshold-reuse and incremental update
• Illustrative example
Proposed Approach: Computational Refinement
1 9 9 92 9 9 93 8 7 63 4 5 5
1 -1 -1 -1-1 -1 -1 -1-1 -1 -1 -1-1 -1 -1 -1
-1 0.6 1 10.60.8 1 11 1 1 11 1 1 1
1 -1 -1 -11 -1 -1 -1-1 -1 -1 -1-1 -1 -1 -1
-.3 0.2 1 1-.6 0.5 1 10.60.8 1 11 1 1 1
(a) feature values (b) indicators, focal values for δ=1 (c) indicators, focal values for δ=2
1 -1 -1 -11 -1 -1 -11 -1 -1 -1-1 -1 -1 -1
-.3 0.2 1 1-.2 .25 1 1-.6 0.5 1 10.30.6 1 1
1 -1 -1 -11 -1 -1 -11 -1 -1 -11 -1 -1 -1
-.3 0.2 1 1-.2 .25 1 1-.2 .25 1 1-.3 0.2 1 1
(d) indicators, focal values for δ=3(intermediate)