Collaborative Learning of Discrete Distributions under Heterogeneity and Communication Constraints

Authors: Xinmeng Huang, Donghwan Lee, Edgar Dobriban, Hamed Hassani

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Further, we provide experimental results using both synthetic data and n-gram frequency estimation in the text domain, which corroborate its efficiency.
Researcher Affiliation Academia Xinmeng Huang , Donghwan Lee Graduate Group in Applied Mathematics and Computational Science University of Pennsylvania Philadelphia, PA 19104 {xinmengh, dh7401}@sas.upenn.edu Edgar Dobriban Department of Statistics and Data Science University of Pennsylvania Philadelphia, PA 19104 dobriban@wharton.upenn.edu Hamed Hassani Department of Electrical and Systems Engineering University of Pennsylvania Philadelphia, PA 19104 hassani@seas.upenn.edu
Pseudocode Yes Algorithm 1 SHIFT: Sparse Heterogeneity Inspired collaboration and Fine-Tuning
Open Source Code Yes Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes]
Open Datasets Yes We test SHIFT on synthetic data as well as the Shakespeare dataset [11]. The Shakespeare dataset was proposed as a benchmark for federated learning in [11].
Dataset Splits No The paper mentions generating synthetic data and drawing datapoints for the Shakespeare dataset, but it does not specify explicit train/validation/test splits for reproduction.
Hardware Specification No The paper states "Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [N/A]", indicating no specific hardware details are provided.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies used in the experiments (e.g., programming language versions, library versions, or solver versions).
Experiment Setup Yes We set the threshold parameter α = ln(n) and the trimming proportion ω = 0.1. We set the dimension to d = 300 and run the simulation by varying n, T, s. We set the uniform distribution, p = (1/d, . . . , 1/d) as the central distribution.