Model Linkage Selection for Cooperative Learning

Authors: Jiaying Zhou, Jie Ding, Kean Ming Tan, Vahid Tarokh

JMLR 2021 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive numerical studies are conducted to assess the performance of the proposed method.
Researcher Affiliation Academia Jiaying Zhou EMAIL School of Statistics University of Minnesota Minneapolis MN, 55455 Jie Ding EMAIL School of Statistics University of Minnesota Minneapolis MN, 55455 Kean Ming Tan EMAIL Department of Statistics University of Michigan Ann Arbor MI, 48109 Vahid Tarokh EMAIL Department of Electrical Engineering and Engineering Duke University Durham NC, 27708
Pseudocode Yes Algorithm 1 Greedy Algorithm for Model Linkage Selection. Input: User-specified graph G, data D(κ), parameter vector θκ, parametric distribution p(κ) θκ ( | θκ, X(κ)), prior distribution πκ( ) on parameters θκ, for κ = 1, . . . , M. 1: Initialize the index ℓ= 1, linkage set ζ(1) = {1}. 2: for ℓ= 2, . . . , M do 3: Let NG(ζ(ℓ 1)) denote the neighboring set of ζ(ℓ 1) within G, namely the set of learners in G\ζ(ℓ 1) that have a model linkage with at least one learner in ζ(ℓ 1). 4: Calculate jopt = argmaxj NG(ζ(ℓ 1))p( κ ζ(ℓ 1)y(κ) | κ ζ(ℓ 1)X(κ), D(j)). 5: if p( κ ζ(ℓ 1)y(κ)| κ ζ(ℓ 1) X(κ)) p( κ ζ(ℓ 1)y(κ) | κ ζ(ℓ 1)X(κ), D(jopt)) break 6: Let ζ(ℓ) = {jopt} ζ(ℓ 1). 7: if ζ(ℓ) = C(G) break 8: end for 9: For a new predictor ex, let bp(ℓ) = p( | κ ζ(ℓ) D(κ), ex) and bπ(ℓ) = π( | κ ζ(ℓ) D(κ)). Output: Predictive distribution bp = bp(ℓ), posterior distribution bπ = bπ(ℓ), and model linkage graph b G.
Open Source Code No The paper does not provide concrete access to source code for the methodology described. It only provides a link to the journal's attribution requirements page, not a code repository.
Open Datasets Yes We consider the Wisconsin Breast Cancer database (Mangasarian et al., 1995).
Dataset Splits Yes We randomly choose 100 samples as the test data for evaluating prediction accuracy. Then, the data are randomly divided into 10 learners, in which each learner has n samples.
Hardware Specification No The paper does not provide specific hardware details used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details, such as library or solver names with version numbers, needed to replicate the experiment.
Experiment Setup Yes We apply Algorithm 1 with the aforementioned user-specified graph and impose a multivariate Gaussian distribution, Np(0, 4Ip), as the prior distribution for the regression coefficients for all learners.