Computing Optimal Monitoring Strategy for Detecting Terrorist Plots

Authors: Zhen Wang, Yue Yin, Bo An

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments showing that our algorithm can obtain a robust enough solution outperforming widely-used centrality based heuristics significantly and scale up to realistic-sized problems.
Researcher Affiliation Academia Zhen Wang School of Computer Engineering Nanyang Technological University Singapore 639798 wangzhen@ntu.edu.sg Yue Yin The Key Lab of Intelligent Information Processing, ICT, CAS University of Chinese Academy of Sciences Beijing 100190, China melody1235813@gmail.com Bo An School of Computer Engineering Nanyang Technological University Singapore 639798 boan@ntu.edu.sg
Pseudocode Yes Algorithm 1: DO-TPD overview; Algorithm 2: better O-D (x, y); Algorithm 3: better O-A (x, y); Algorithm 4: Local Search (v, A, x)
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets Yes We conduct experiments on three types of graph structures which are widely used to model terrorist organisations: (i) Random trees (RT), where every new vertex is attached to a randomly picked incumbent; (ii) Erd os-R enyi random graphs (ER(V , M)), where exactly M edges are randomly constructed between all the possible pairs of vertices (Erd os and R enyi 1959); (iii) Barab asi-Albert scale-free networks (BA(k)), where each new vertex is connected to k incumbents using a preferential attachment mechanism (Barab asi and Albert 1999). We conduct experiments on two 9/11 networks (a small one with 19 vertices, who are directly responsible for this attack, and a bigger one with 37 vertices including accomplices (Krebs 2002))
Dataset Splits No The paper does not explicitly provide details about training, validation, or test dataset splits. The problem is formulated as a game, and the algorithms find optimal strategies, rather than training a model on data splits.
Hardware Specification Yes All computations are performed on a machine with a 3.20GHz quad core CPU and 16.00GB memory.
Software Dependencies Yes All LPs and MILPs are solved with CPLEX (version 12.6).
Experiment Setup Yes The parameters in better O-A (Algorithm 3), tmax, cmax and kmax are set to |V |, 0.2 |V | and 3, respectively. The number of resources R is set to |V | / 5 unless otherwise specified. The capability of each vertex τv is randomly chosen in [1, 5], and the network externality measure δ is fixed to be 0.1.