REPrune: Channel Pruning via Kernel Representative Selection

Authors: Mincheol Park, Dongjin Kim, Cheonjun Park, Yuna Park, Gyeong Eun Gong, Won Woo Ro, Suhyun Kim

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results highlight that REPrune performs better in computer vision tasks than existing methods, effectively achieving a balance between acceleration ratio and performance retention.
Researcher Affiliation Collaboration 1Yonsei University, Republic of Korea 2Korea University, Republic of Korea 3Korea Institute of Science and Technology, Republic of Korea 4Hyundai MOBIS, Republic of Korea
Pseudocode Yes Algorithm 1: Channel Selection in REPrune; Algorithm 2: Overview of REPrune
Open Source Code No The paper mentions using the PyTorch framework but does not provide a link to their own source code for REPrune.
Open Datasets Yes Datasets and models We evaluate image recognition on CIFAR-10 and Image Net (Deng et al. 2009) datasets and object detection on COCO-2017 (Lin et al. 2014).
Dataset Splits Yes Image Net validation dataset.
Hardware Specification Yes While evaluations are conducted on NVIDIA RTX A6000 with 8 GPUs, for the CIFAR-10 dataset, we utilize just a single GPU. Additionally, we employ NVIDIA Jetson TX2 to assess the image throughput of our pruned model with A6000.
Software Dependencies No The paper states: "This paper evaluates the effectiveness of REPrune using the Py Torch framework", but it does not specify any version numbers for PyTorch or other software dependencies.
Experiment Setup No The paper states: "Comprehensive details regarding training hyper-parameters and pruning strategies for CNNs are available in the Appendix.", indicating that these details are not provided in the main text.