Graph Algorithm – Efficient Shortest Path Estimation Mentee: Yonk Shi (CSE, Moorpark College) Mentor: Arijit Khan (CS, UCSB) Faculty Advisor: Dr. Xifeng Yan Computer Science Department of UCSB INSET Program 1
Shortest Path Algorithm A B Simple case: We are trying to develop a general • algorithm for graph navigation Reality: It will work with any dataset, i.e. Google, • Facebook, Last.fm It is optimized for massive databases • It is extremely efficient regardless of the • size of the graph. Source: my facebook, www.facebook.com/yonkshi
Algorithm Dimensionless Data 2 Dimensional Data S1 S2 A S1 B S2 A B MDS: Multidimensional Scaling • Preserved Distances • Preserved Paths
Algorithm In reality, MDS generates an approximation of coordinates, thus the distance is approximated
Experimental Results Distances Calculated by Different Algorithms 6 5 4 MDS-distance Distance MDS-Dij-Distance 3 Real DIjsktra Distance Our Algorithm 2 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
Experimental Results Steps Taken by Different Algorithms 7 6 5 4 Steps Our Algorithm 3 Dijsktra 2 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 10% Failure Rate
Experimental Results Average Time Taken by Algorithms 0 5 10 15 20 25 30 35 Our Algorithm Dijkstra's As much as 3000x Faster than Dijkstra’s Algorithm
Conclusion • We have designed a shortest path algorithm • It is very efficient and accurate for large databases • It is much faster than Dijkstra’s Algorithm Our future goals: • Reduce failure rate to 0% (while maintaining accuracy) • Increase high efficiency and accuracy • Add “Label” information for even more accurate search
Thank You! Special Thanks To: My mentor Arijit Khan Faculty Advisor Dr Xifeng Yan Professor Christine Aguilera from Moorpark College Professor Martin Chetlen from Moorpark College Awesome Roommates Adam, Michael and Jose
1 0.8 0.6 0.4 Path 0.2 1340 0 -1 -0.5 0 0.5 1 -0.2 -0.4 -0.6
Recommend
More recommend
