Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Approximate Gradient Coding for Distributed Learning with Heterogeneous Stragglers
Authors: Heekang Song, Wan Choi
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we demonstrate the effectiveness of the proposed optimally structured gradient coding scheme for straggler mitigation in distributed learning. We numerically evaluate its performance on the large-scale COCO dataset [21]. |
| Researcher Affiliation | Academia | Heekang Song Korea Advanced Institute of Science and Technology School of Electrical Engineering EMAIL Wan Choi Seoul National University Department of Electrical and Computer Engineering EMAIL |
| Pseudocode | Yes | Algorithm 1 The Sparse Code Construction Algorithm Calculate Y1, Y2, ..., Yk. Y {Y1, Y2, ..., Yk} and Y = . while |Y | = |Y| do |
| Open Source Code | Yes | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: We provide the code for our paper and its instructions in anonymous github repository. |
| Open Datasets | Yes | In this section, we demonstrate the effectiveness of the proposed optimally structured gradient coding scheme for straggler mitigation in distributed learning. We numerically evaluate its performance on the large-scale COCO dataset [21]. |
| Dataset Splits | Yes | Throughout the experiments, the loss represents the overall training objective of the object-detection model the sum of classification and bounding-box regression losses computed on the COCO validation set. |
| Hardware Specification | Yes | The experiments are conducted on one NVIDIA Ge Force RTX 3060 GPU (12 GB), six NVIDIA Ge Force GTX 1080 GPUs (8 GB each), and twelve NVIDIA Tesla P100 GPUs (16 GB each) (provided through Kaggle Cloud). |
| Software Dependencies | Yes | Problem (P3) is a convex problem with respect to the transformed variables αj i and can be solved by the standard convex optimization tool, such as CVX [15] and YALMIP [16]. [15] M. Grant and S. Boyd, CVX: Matlab software for disciplined convex programming, v2.1, Mar. 2014. [Online]. Available: http://cvxr.com/cvx |
| Experiment Setup | Yes | In our experiments, we employ the Mobile Net V3 model, and the learning rate is set to γt = 0.01. Suppose τth denote the response time limit for each training iteration. ... We set k = 10, ψmin = 0.1, ψmax = 2, and τth = 1.1, unless stated otherwise. |