Tiny Déjà Vu: Smaller Memory Footprint & Faster Inference on Sensor Data Streams with Always-On Microcontrollers

This repository includes the core implementation parts of Tiny Déjà Vu.

Reproduce Experimental Results on MCU

The experiments were conducted via RIOT-ML. To reproduce the results it is required to get RIOT-ML ready.

0. Go get a IoT board e.g. STM32 nucleo-f767zi. Connect the IoT board to your PC.
1. Clone the RIOT-ML from https://github.com/TinyPART/RIOT-ML/.
2. Follow the Prequisites section, install all necessary packages and toolchains.
3. Download the generated C-Code from here, and extract it to the to the RIOT-ML directory.
4. Copy eval_ssm_HIL.py and main.c to the RIOT-ML directory.
5. Run python eval_ssm_HIL.py under the RIOT-ML directory.
6. Grab some coffee and wait for the results written in SSM_eval_result_{board}.json :).

It is noted that r in this repo / results actually represents $1-r_{overlap}$.