MCUNet: Tiny Deep Learning on IoT Devices
Authors: Ji Lin, Wei-Ming Chen, Yujun Lin, john cohn, Chuang Gan, Song Han
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We used 3 datasets as benchmark: Image Net [14], Visual Wake Words (VWW) [12], and Speech Commands (V2) [46]. We deploy the models on microcontrollers of diverse hardware resource, including STM32F412 (Cortex-M4, 256k B SRAM/1MB Flash), STM32F746 (Cortex-M7, 320k B/1MB Flash), STM32F765 (Cortex-M7, 512k B SRAM/1MB Flash), and STM32H743 (Cortex-M7, 512k B SRAM/2MB Flash). By default, we use STM32F746 to report the results unless otherwise specified. |
| Researcher Affiliation | Collaboration | Ji Lin1 Wei-Ming Chen1,2 Yujun Lin1 John Cohn3 Chuang Gan3 Song Han1 1MIT 2National Taiwan University 3MIT-IBM Watson AI Lab |
| Pseudocode | No | The paper describes algorithmic steps in prose but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper lists 'https://tinyml.mit.edu' on the first page, which is a project website. It also references '[1] Solution to visual wakeup words challenge 19 (first place). https://github.com/mit-han-lab/VWW.' However, there is no unambiguous statement explicitly stating that the source code for the MCUNet methodology described in this paper is released or available at a specific repository. |
| Open Datasets | Yes | We used 3 datasets as benchmark: Image Net [14], Visual Wake Words (VWW) [12], and Speech Commands (V2) [46]. |
| Dataset Splits | Yes | During neural architecture search, in order not to touch the validation set, we perform validation on a small subset of the training set (we split 10,000 samples from the training set of Image Net, and 5,000 from VWW). Speech Commands has a separate validation&test set, so we use the validation set for search and use the test set to report accuracy. |
| Hardware Specification | No | The paper mentions 'MIT Satori cluster for providing the computation resource' but does not specify the exact hardware components (e.g., GPU models, CPU models, memory) of this cluster used for running the experiments. It lists microcontroller specifications for deployment, not for training/experimentation setup. |
| Software Dependencies | No | The paper mentions software frameworks like TensorFlow Lite Micro, CMSIS-NN, TensorFlow, and PyTorch, but it does not provide specific version numbers for any of these or other ancillary software dependencies. |
| Experiment Setup | No | The paper states 'The training details are in the supplementary material', meaning explicit hyperparameters or detailed training configurations are not provided within the main text. |