Optimally Improving Cooperative Learning in a Social Setting
Authors: Shahrzad Haddadan, Cheng Xin, Jie Gao
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The performance of all of our algorithms are guaranteed by mathematical analysis and backed by experiments on synthetic and real data.Our experiments show that by modifying only a few vertices, we succeed in increasing the accuracy of a high percentage of network s agents; see Figure 1. |
| Researcher Affiliation | Academia | Shahrzad Haddadan 1 Cheng Xin 2 Jie Gao 2 1Rutgers Business School, Piscataway, NJ, USA 2Department of Computer Science, Rutger University, Piscataway, NJ, USA. Correspondence to: Shahrzad Haddadan <shaddadan@business.rutgers.edu>, Jie Gao <jg1555@rutgers.edu>. |
| Pseudocode | Yes | A.3. Pseudocode of the greedy algorithms In this section we present our algorithms for egalitarian improvement. Algorithm 1 Egal Alg π, W... Algorithm 2 Egal Alg(appx) mode, err, W... Algorithm 3 Est Gainind(S, u, {err(vj)}n j=1, W)... Algorithm 4 Est Gaingr(S, u, {err(vj)}n j=1, err(R), err(B), W)... |
| Open Source Code | Yes | Code of our experiments is available through link 3. 3https://github.com/jackal092927/social_learning_public |
| Open Datasets | Yes | Synthetic Datasets Synthetic data is generated with three components: a random graph G, a weight matrix W and initial opinions ˆy. To generate G, we employ Erd os-Rényi model (ER) (Erdös & Rényi, 1959), Barabási-Albert model (PA) (Barabási & Albert, 1999) and Watts-Strogatz model (WS) (Watts & Strogatz, 1998). Real-World Graphs We also evaluate our methods on four diverse real network datasets (Rossi & Ahmed, 2015), BIO (Duch & Arenas, 2005; Bader et al., 2012), CSPK (Bader et al., 2013), FB (Rozemberczki et al., 2019), WIKI (Leskovec et al., 2010). |
| Dataset Splits | No | The paper does not explicitly provide training/test/validation dataset splits (e.g., percentages, counts, or references to predefined splits). It mentions evaluating on datasets and calculating expected values by taking averages over Ω, but no detailed partitioning for reproducibility of splits. |
| Hardware Specification | No | The paper does not explicitly describe the hardware used for running the experiments. There are no mentions of specific GPU models, CPU types, or other detailed hardware specifications. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers. It mentions generating synthetic data and using certain models but does not list the software environment with versions necessary for replication. |
| Experiment Setup | Yes | B. Additional experiments and details of set up... Number of nodes: 128 Erd os-Rényi graph: Probability for edge creation p = 0.005. Barabási-Albert preferential attachment model: Number of edges to attach from a new node to existing nodes m = 5 Watts-Strogatz small-world graph: Each node is joined with its k = 5 nearest neighbors in a ring topology; The probability of rewiring each edge p = 0.25... We let |Ω| = 3 and for initial opinion ˆy, we randomly generate for each agent vi a random vector (ˆyi(a) : a Ω) with each entry sampled from Bernoulli distribution with probability pi of ˆyi(a) = +1 sampled uniformly from [0.3, 0.9]. |