TRQ: Ternary Neural Networks With Residual Quantization
Authors: Yue Li, Wenrui Ding, Chunlei Liu, Baochang Zhang, Guodong Guo8538-8546
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results demonstrate that the proposed method yields great recognition accuracy while being accelerated. |
| Researcher Affiliation | Collaboration | Yue Li1, Wenrui Ding1*, Chunlei Liu1, Baochang Zhang1, Guodong Guo2 1 Beihang University 2 Institute of Deep Learning, Baidu Research and National Engineering Laboratory for Deep Learning Technology and Application |
| Pseudocode | No | The paper describes the methodology using mathematical formulas and prose, but it does not include any explicitly labeled pseudocode blocks or algorithm sections. |
| Open Source Code | No | The paper does not contain an unambiguous statement that the authors are releasing their code for the described methodology, nor does it provide a direct link to a source-code repository. |
| Open Datasets | Yes | We perform diverse experiments on three classification datasets: CIFAR-10/100 (Alex Krizhevsky 2014) and Image Net (ILSVRC 2012) (Russakovsky et al. 2015). |
| Dataset Splits | Yes | For Image Net, training images are randomly cropped into the resolution of 224 224. After that, the images are normalized using the mean and standard deviation. No additional augmentations are performed except the random horizontal flip. However, for validation images, we use center crop instead of random crop and no flip is applied. ... (b) Validation accuracy curves on Image Net. |
| Hardware Specification | No | The paper discusses computational complexity but does not specify the exact GPU models, CPU models, or other detailed hardware specifications used to run the experiments. |
| Software Dependencies | No | The paper mentions using 'Adam with momentum of 0.9' as an optimizer but does not specify version numbers for any software libraries, frameworks, or programming languages used for implementation. |
| Experiment Setup | Yes | For CIFAR-10/100, we run the training algorithm for 200 epochs with a batch size of 256. Besides, a linear learning rate decay scheduler is used, and the initial learning rate is set to 0.01. For experiments on Image Net, we train the models for up to 100 epochs with a batch size of 256. The learning rate starts from 0.001 and is decayed twice by multiplying 0.1 at 75th and 95th epoch. For all settings, Adam with momentum of 0.9 is adopted as the optimizer. |