Towards Accurate Low Bit-Width Quantization with Multiple Phase Adaptations
Authors: Zhaoyi Yan, Yemin Shi, Yaowei Wang, Mingkui Tan, Zheyang Li, Wenming Tan, Yonghong Tian6591-6598
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that MPA achieves higher accuracy than most existing methods on classification tasks for Alex Net, VGG-16 and Res Net. |
| Researcher Affiliation | Collaboration | Zhaoyi Yan,1 Yemin Shi,2 Yaowei Wang,3 Mingkui Tan,3 Zheyang Li,4 Wenming Tan,4 Yonghong Tian2 1School of ECE, Peking University Shenzhen Graduate School 2School of EECS, Peking University, 3Pengcheng Laboratory, 4Hikvision Research Institute |
| Pseudocode | Yes | Algorithm 1 Training a L-layer network with k cluster center weights Multiple Phase Adaptations Quantization |
| Open Source Code | No | The paper does not contain any explicit statements about releasing source code, providing a link to a repository, or indicating that code is available in supplementary materials. |
| Open Datasets | Yes | We use the Res Net-18 in order to conduct quantization on CIFAR-10 to prove the effectiveness of our method. |
| Dataset Splits | Yes | CIFAR-10 consists of 10 classes with 6,000 images per class. Each image is a 32 32 color picture. There are 50,000 training images and 10,000 test images. |
| Hardware Specification | No | The paper only states 'We use GPU cards for the experiments with a batch size of 256.' which does not provide specific hardware details such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper states 'The deep learning framework that we use is Py Torch.' but does not specify a version number for PyTorch or any other software components, which is required for reproducible dependency information. |
| Experiment Setup | Yes | The RMSprop optimizer is adopted with an initial learning rate of 0.005 and a momentum of 0.5. [...] We set λ0 = 0.01 and d0 = 0.95. [...] Quantization is conducted with a learning rate of 0.001 and other parameters unchanged. |