Pruning via Sparsity-indexed ODE: a Continuous Sparsity Viewpoint

Authors: Zhanfeng Mo, Haosen Shi, Sinno Jialin Pan

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical experiments show that PSO achieves either better or comparable performance compared to state-of-the-art baselines across various pruning settings.
Researcher Affiliation Collaboration 1School of Computer Science and Engineering, NTU, Singapore 2Continental-NTU Corporate Lab, NTU Singapore 3Department of Computer Science and Engineering, Chinese University of Hong Kong.
Pseudocode Yes Algorithm 1 Pruning via Sparsity-indexed ODE (PSO)
Open Source Code Yes Our implementations are now available on Git Hub1. 1https://github.com/mzf666/sparsity-indexed-ode
Open Datasets Yes CIFAR-10 / 100 benchmarks (Krizhevsky, 2009) using different model architectures (Res Net-20 (He et al., 2016), VGG16-bn (Simonyan & Zisserman, 2015) and WRN-20 (Zagoruyko & Komodakis, 2016))
Dataset Splits No The paper uses standard public datasets (e.g., CIFAR-10/100, ImageNet) which have predefined train/test splits. However, it does not explicitly provide details about a specific validation set split (e.g., percentage or sample count) used during training or fine-tuning.
Hardware Specification No The paper mentions running experiments on various datasets and models but does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used for these experiments.
Software Dependencies No The paper states that implementations are available on GitHub but does not list specific software dependencies with their version numbers within the text.
Experiment Setup Yes Table 6. Hyperparameter configurations for retraining procedure. VGG16-bn...Optimizer SGD-Momentum Training Epochs 100 Batch Size 64 Learning Rate 1e-2 Learning Rate Schedule Cosine Annealing Minimal Learning Rate and When to Reach 1e-6 last 10 epochs Evaluated Epochs last 10 epochs