個人や組織の活動に伴って時系列的に発生する情報をサーバで安全に集計することは実世界で大きなニーズがある.この集計において,サーバは受け取る時系列情報の範囲を予め予想できないめ,情報の受取りに伴って集計表への加算だけではなく,集計表の拡張を行う必要がある.本論文では,時系列情報の安全な集計問題を新たに定義した上で,秘密分散によって時系列情報を秘匿しながらマルチパーティ計算によって集計する4 つの方式を提案する.メモリ使用量,通信ラウンド数,通信量の評価により,4つのうち表を全探索する方式はメモリ容量とラウンド数が少なく,再帰的PathORAM を用いる方式は通信量が少ないことを明らかにする.
People often need to use servers to count on time-series information that is generated during
activities of people and organizations. In this counting, because a server cannot predict range of information
to accept in future, a server needs not only to add numbers on the counting table but also to extend the table.
This paper provides new denition of secure counting on time-series information. Based on this denition,
the paper describes four methods that hide time-series information by secret sharing and that count on it by
multiparty computation. These methods are evaluated in terms of memory size, communication round, and
communication amount. It is claried that a method that accesses the table exhaustively is best in memory
size and communication round and a method that uses recursive Oblivious Random Access Machine is best
in communication amount.
A Distributed Parallel Community Detection Heuristics for Large-scale Real Graphs
1. Symposia on VLSI Technology and Circuits
A Distributed Parallel
Community Detection Heuristics
for Large-scale Real Graphs
Guanglong Chi1,a,b
Ken Wakita2,a,b
Hitoshi Sato3,a,b
1 Department of Human System Science
2 Department of Mathematical and Computing Science
3 Global Scientific Information and Computing Center
a Tokyo Institute of Technology
b JST, CREST
10. 既存の分散手法:PMETISNC-Louvain
(Wickramaarachchi+, HPEC `14)
? 前処理:
PMETISグラフノード分割
(Karypis+, SIAM `98)
? 理由:
– ノード分割により、
計算ノードに跨る
コミュニティ数が少なくなる
– 計算ノード間の通信が
無視できる
? Wickramaarachchi, C., Frincu, M., Small, P. and Prasanna, V. K.: Fast parallel algorithm for
unfolding of communities in large graphs, IEEE HPEC `14, pp.1–6.
? Karypis, G. and Kumar, V.: A fast and high quality multilevel scheme for partitioning irregular
graphs, SIAM Journal on scientific Computing, Vol. 20, No. 1, pp. 359–392 (1998).
#11: PMETISNC-Louvain: PMETIS without communication Louvain
PMETISC-Louvain: PMETIS with communication Louvain
ICC-Louvain: IC with communication Lovuain