Exploiting Anonymity in Approximate Linear Programming: Scaling to Large Multiagent MDPs

Authors: Philipp Robbel, Frans Oliehoek, Mykel Kochenderfer

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

Reproducibility Variable Result LLM Response
Research Type Experimental 6 Experimental Evaluation We evaluate our methods on undirected disease propagation graphs with 30 and 50 nodes. For the first round of experiments, we contrast runtimes of the normal VE/ALP method (where possible) with those that exploit anonymous influence . We then consider a disease control problem with 25 agents in a densely connected 50-node graph that cannot be solved with the normal ALP.
Researcher Affiliation Academia Philipp Robbel MIT Media Lab Cambridge, MA, USA Frans A. Oliehoek University of Amsterdam University of Liverpool Mykel J. Kochenderfer Stanford University Stanford, CA, USA
Pseudocode No The paper describes methods like AUGMENT and REDUCE in prose and mathematical notation within Section 4 but does not present them in a structured pseudocode or algorithm block format.
Open Source Code No The paper does not provide any explicit statement about releasing its source code or a link to a code repository for the methodology described.
Open Datasets No We use graph-tool (Peixoto 2014) to generate 10 random graphs with an out-degree k sampled from P(k) 1/k, k [1, 10]. The paper describes how the graphs were generated, but does not provide concrete access information (link, DOI, specific citation) for the generated datasets themselves.
Dataset Splits No The paper evaluates mean returns from 50 randomly sampled starting states after 200 steps of policy simulation, but it does not provide specific training/validation/test dataset splits (e.g., percentages or counts) or refer to standard predefined splits.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments, such as GPU/CPU models, memory, or cloud computing specifications.
Software Dependencies No The paper mentions using 'graph-tool (Peixoto 2014)' for generating graphs, but it does not provide specific version numbers for software dependencies or libraries used in the experimental setup beyond this citation.
Experiment Setup Yes In all experiments, we use indicator functions IXi, I Xi on each state variable (covering the two valid instantiations {healthy, infected}) as the basis set H in the (RR-)ALP. We use identical transmission and node recovery rates throughout the graph, β = 0.6, δ = 0.3. Action costs are set to λ1 = 1 and infection costs to λ2 = 50. All experiments use the identical greedy elimination heuristic for both VE and RR-VE, which minimizes the scope size of intermediate terms at the next iteration.