The Influence of Memory in Multi-Agent Consensus
Authors: David Kohan Marzagão, Luciana Basualdo Bonatto, Tiago Madeira, Marcelo Matheus Gauy, Peter McBurney11254-11262
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We provide a theoretical analysis of the probability of each option eventually winning such processes based on the initial opinions expressed by agents. Further, we perform experiments to investigate network topologies in which agents benefit from memory on the expected time needed for consensus. We use a mix of probabilistic analysis and simulation to explore these questions. In this section we investigate, through simulations, how the duration (measured in number of round until consensus) of early 1-memory process compares to their memoryless counterparts. We perform two experiments on the network topologies described above. Results of Experiment 1 are shown in Figure 3 with x-axis indicating the different values of p0, whereas the values on the y-axis represent τ. |
| Researcher Affiliation | Academia | David Kohan Marzag ao,1 Luciana Basualdo Bonatto,2 Tiago Madeira,3 Marcelo Matheus Gauy,3 Peter Mc Burney1 1 King s College London 2 University of Oxford 3 University of S ao Paulo |
| Pseudocode | No | The paper does not contain any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | Code and Data The repository containing the code, data, and plots associated to this project can be found at https://github.com/tmadeira/consensus. |
| Open Datasets | No | The paper describes generating graph structures (cliques, cycles, bicliques, binary trees, grids on a torus) for simulations, rather than using a pre-existing publicly available dataset. It states: "For a given pair (p0, n) and a graph type, we denote the the ratio between the average consensus time of the 1-memory process and the average consensus time of its memoryless counterpart by τ." |
| Dataset Splits | No | The paper describes generating graph structures and initial conditions for simulations rather than using a train/validation/test split on a pre-existing dataset. It mentions: "Each experiment compares a memoryless process with a given initial configuration with its early 1-memory counterpart with the same initial state... To avoid bias given by the initial state, each set of experiments (with and without memory) has a different random starting point, with each node being red or blue with equal probability." |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments. It only mentions that simulations were performed. |
| Software Dependencies | No | The paper does not list specific software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x) that are needed to replicate the experiments. |
| Experiment Setup | Yes | When considering the 1-memory consensus process (p0, p1) on undirected graph G, we assume that, with probability p0 (resp. p1), a node copies the present (resp. past) colour of a neighbour chosen uniformly at random. In the first experiment, we fix the number of nodes n, while varying p0 to explore the effect of memory for these different values. The second experiment fixes a value of p0 to investigate how improvement of memory changes with n. For a given pair (p0, n) and a graph type, we denote the the ratio between the average consensus time of the 1-memory process and the average consensus time of its memoryless counterpart by τ. We perform two experiments on the network topologies described above. In experiment 1, we have recorded the duration of 4000 simulations for graphs of size n = 1023, for 30 different values of p0, ranging uniformly from 0.1 to 1. The setup is analogous to the one in Experiment 1, with the difference that we now average over 104 simulations for each n and each graph type. The values chosen for n depend on the type of graph. |