MobileTL: On-Device Transfer Learning with Inverted Residual Blocks
Authors: Hung-Yueh Chiang, Natalia Frumkin, Feng Liang, Diana Marculescu
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on multiple datasets demonstrate that our method is Pareto-optimal (best accuracy under given hardware constraints) compared to prior work in transfer learning for edge devices. |
| Researcher Affiliation | Academia | Hung-Yueh Chiang, Natalia Frumkin, Feng Liang, Diana Marculescu The University of Texas at Austin Chandra Family Department of Electrical and Computer Engineering |
| Pseudocode | No | The paper describes its methods verbally and through diagrams but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any statement or link regarding the public release of its source code. |
| Open Datasets | Yes | We apply transfer learning on multiple image classification tasks. Similar to prior work (Kornblith, Shlens, and Le 2019; Houlsby et al. 2019; Cai et al. 2020), we begin with an Image Net (Deng et al. 2009) pre-trained model, and transfer to eight downstream image classification datasets, including Cars (Krause et al. 2013), Flowers (Nilsback and Zisserman 2008), Aircraft (Maji et al. 2013), CUB-200 (Wah et al. 2011), Pets (Parkhi et al. 2012), Food (Bossard, Guillaumin, and Gool 2014), CIFAR10 (Krizhevsky, Hinton et al. 2009), and CIFAR100 (Krizhevsky, Hinton et al. 2009). |
| Dataset Splits | No | The paper mentions using well-known datasets like CIFAR10 and CIFAR100, which have standard splits, but it does not explicitly state the specific training, validation, or test dataset split percentages or sample counts used for reproduction. It only mentions 'train for 50 epochs with a batch size of 8'. |
| Hardware Specification | Yes | We experiment with Raspberry PI4 model B with Quad core ARM Cortex-A72 64-bit and 4 GB RAM, and NVIDIA JETSON NANO with 128-core GPU, Quadcore ARM Cortex-A57, and 2 GB RAM. |
| Software Dependencies | No | The paper mentions using the 'Py Torch framework' but does not specify its version or any other software dependencies with version numbers. |
| Experiment Setup | Yes | For fair comparison, we follow the hyper-parameters and settings in Tiny TL (Cai et al. 2020) where they train for 50 epochs with a batch size of 8 on a single GPU. We use the Adam optimizer and cosine annealing for all experiments, however, the initial learning rate is slightly tuned for each dataset and model. The classification layers are trained in all settings, and fusion layers are trained in block-wise fine-tuning. We ran our experiment using four random seeds, and average the results. |