SASB: S patial A ctivity S ummarization using B uffers Atanu Roy & Akash Agrawal
Feb 22, 2016
SASB: Spatial Activity Summarization using Buffers
Atanu Roy & Akash Agrawal
Overview• Motivation• Problem Statement• Computational Challenges• Related Works• Approach• Examples• Conclusion
Motivation
• Applications in domains like
– Public safety
– Disaster relief operations
SASB
SASB Problem Statement• Input
– A spatial network,– Set of activities & their location in space,– Number of buffers required (k),– A set of buffer (β),
• Output– A set of k active buffers, where
• Objective– Maximize the number of activities covered in the k buffers
• Constraints– Minimize computation costs
Definitions
• Constant Area Buffers– Node buffers
– Path buffers
Running Example
CoveragePath Buffer = 16 Node Buffer = 15 Total Coverage = 31/33
Computational Challenges
• SASB is NP-Hard• Proof:
– KMR is a special case of SASB • Buffers have width = 0
– KMR is proved to be NP-Complete– SASB is at least NP-Hard
Related WorksGeometry based
No Yes
Network based
YesPath based:
KMR, Mean Streets
0-1 Subgraph:SANET, Max Subgraph
This work
No-
K-Means, K-Medoids, P-median,
Hierarchical Clustering
Contributions
• Definition SASB problem • NP-Hardness proof• Combination of geometry and network based
summarization.• First principle examples
Greedy Approach
Choice of k-best buffers
• Repeat k times– Choose the buffer with maximum activities – Delete all activities contained in the chosen buffer from all the
remaining buffers– Replace the chosen buffer from buffer pool to the result-set
Execution Trace
NB_A = 8
NB_B = 6
NB_C = 11
PB_1 = 8
PB_2 = 12
PB_12 = 2
NB_A = 8 NB_A = 8
NB_B = 6 NB_B = 2
NB_C = 11 NB_C = 1
PB_1 = 8 PB_1 = 7
PB_2 = 12 PB_2 = NA
PB_12 = 2 PB_12 = 1
Execution Trace: Final Solution
Best Case Scenario
Type Coverage Rank
Total 33 NA
Geo 27 2
N/w 14 3
SASB 31 1
Better
Type Coverage Rank
Total 14 NA
Geo 9 3
N/w 11 2
SASB 14 1
Average Case Scenario
Type Coverage
Rank
Geo 10 3
N/w 10 2
SASB
1
Type Coverage Rank
Total 10 NA
Geo 8 2
N/w 10 1
SASB 10 1
Conclusion
• Provides a framework to fuse geometry and network based approaches.
• First principle examples indicates it can be comparable with related approaches.
Acknowledgements
• CSci 8715 peer reviewers who gave valuable suggestions.
Thank you