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..

DropNet: Reducing Neural Network Complexity via Iterative Pruning

Authors: Chong Min John Tan, Mehul Motani

ICML 2020 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we show that Drop Net is robust across diverse scenarios, including MLPs and CNNs using the MNIST, CIFAR-10 and Tiny Image Net datasets.
Researcher Affiliation Academia 1Department of Electrical and Computer Engineering, National University of Singapore.
Pseudocode Yes Algorithm 1 Iterative Pruning Algorithm
Open Source Code Yes To encourage further research on iterative pruning techniques, the source code used for our experiments is publicly available at https://github.com/tanchongmin/Drop Net.
Open Datasets Yes we test it empirically using MLPs and CNNs on MNIST (Le Cun et al., 2010), CIFAR-10 (Krizhevsky et al., 2009) and Tiny Image Net (taken from https://tinyimagenet.herokuapp.com, results in Supplementary Material) datasets.
Dataset Splits Yes For MNIST, the dataset is split into 54000 training, 6000 validation and 10000 testing samples. For CIFAR-10, the dataset is split into 45000 training, 5000 validation and 10000 testing samples.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions using "off-the-shelf machine learning libraries" but does not specify any software names with version numbers required for replication.
Experiment Setup Yes The optimization function used is SGD with a learning rate of 0.1. Training Cycles: The masks are applied at the start of each training cycle, which comprises 100 epochs, with early stopping using validation loss with patience of 5 epochs. Over each training cycle, a fraction p = 0.2 of the nodes are dropped.