SpArSe: Sparse Architecture Search for CNNs on Resource-Constrained Microcontrollers
Authors: Igor Fedorov, Ryan P. Adams, Matthew Mattina, Paul Whatmough
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate that it is possible to automatically design CNNs which generalize well, while also being small enough to fit onto memory-limited MCUs. Our Sparse Architecture Search method combines neural architecture search with pruning in a single, unified approach, which learns superior models on four popular IoT datasets. |
| Researcher Affiliation | Collaboration | Igor Fedorov Arm ML Research igor.fedorov@arm.com Ryan P. Adams Princeton University rpa@princeton.edu Matthew Mattina Arm ML Research matthew.mattina@arm.com Paul N. Whatmough Arm ML Research paul.whatmough@arm.com |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. It references third-party open-source tools it used but does not release its own implementation. |
| Open Datasets | Yes | We report results on a variety of datasets: MNIST (55e3, 5e3, 10e3) [38], CIFAR10 (45e3, 5e3, 10e3) [34], CUReT (3704, 500, 1408) [57], and Chars4k (3897, 500, 1886) [16]... |
| Dataset Splits | Yes | We report results on a variety of datasets: MNIST (55e3, 5e3, 10e3) [38], CIFAR10 (45e3, 5e3, 10e3) [34], CUReT (3704, 500, 1408) [57], and Chars4k (3897, 500, 1886) [16], corresponding to classification problems with 10, 10, 61, and 62 classes, respectively, with the training/validation/test set sizes provided in parentheses. |
| Hardware Specification | Yes | Experiments were run on four NVIDIA RTX 2080 GPUs. ... For validation, we use u Tensor [7] to convert CNNs from Sp Ar Se into baremetal C++, which we compile using mbed-cli [3] and deploy on the Micro Bit and STM32F413 MCUs. |
| Software Dependencies | No | The paper mentions using 'u Tensor [7]' and 'mbed-cli [3]' for deployment and compilation, but it does not specify version numbers for these software components. |
| Experiment Setup | No | The paper describes the multi-objective optimization problem and the stages of the Sp Ar Se framework, along with pre-processing steps ('mean subtraction and division by the standard deviation'). It mentions 'hyperparameters governing the training process' and 'pruning hyperparameters' and refers to the Appendix for 'a complete description of the search space', but specific hyperparameter values (e.g., learning rate, batch size, number of epochs, optimizer settings) are not provided in the main text. |