BATUDE: Budget-Aware Neural Network Compression Based on Tucker Decomposition
Authors: Miao Yin, Huy Phan, Xiao Zang, Siyu Liao, Bo Yuan8874-8882
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experimental results on Image Net dataset show that our method enjoys 0.33% top-5 higher accuracy with 2.52 less computational cost as compared to the uncompressed Res Net-18 model. |
| Researcher Affiliation | Collaboration | Miao Yin1*, Huy Phan1, Xiao Zang1, Siyu Liao2 , Bo Yuan1 1Department of Electrical and Computer Engineering, Rutgers University, 2Amazon {miao.yin, huy.phan, xiao.zang}@rutgers.edu, liasiyu@amazon.com, bo.yuan@soe.rutgers.edu |
| Pseudocode | Yes | Algorithm 1: Budget-Aware Training with BC-ADL |
| Open Source Code | No | The paper does not provide any explicit statements or links indicating the release of open-source code for the described methodology. |
| Open Datasets | Yes | We evaluate the proposed BATUDE for various popular CNN models on CIFAR-10, CIFAR-100 and Image Net datasets. |
| Dataset Splits | No | The paper mentions using CIFAR-10, CIFAR-100, and ImageNet datasets, but does not provide specific details on the training, validation, and test splits (e.g., percentages or sample counts). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. |
| Software Dependencies | No | The paper mentions 'Training optimizer is set as momentum SGD' but does not specify software versions for libraries, frameworks, or programming languages used. |
| Experiment Setup | Yes | Learning rate is set as 0.1 and 0.01 for CIFAR and Image Net dataset, respectively. The learning rate is multiplied by a factor of 0.1 every 30 epochs. Training optimizer is set as momentum SGD. For other hyper-parameters, batch size, weight decay and λ are set as 256, 0.0001, and 0.001. Also, the compressed Tucker-format models are fine-tuned with learning rate 0.005 and 0.001 on CIFAR and Image Net dataset, respectively. |