Post by Rishi Latchmepersad

Happy to share that Tevin Achong and I have published our first research paper on Machine Learning entitled: On-device Temperature Forecasting for Microclimates Using Compressed Neural Networks. 🎉 In this paper, we discuss our findings around forecasting outdoor air temperature using neural networks (NNs) deployed on a small, offline STM32 microcontroller. We initially started off learning the basics of I2C and cheap embedded sensors and ended up deploying and benchmarking multiple NNs. In a nutshell, we learnt that: 1) Baseline models are usually pretty good and hard to improve on. 2) NNs need a lot more data that we imagined, even tiny ones. 3) Ablation models are important to show the impact of different parts of the models. I have shared some images of the setup below. Thanks to Kit Fai Pun and the team at the IEMJ for encouraging us to publish this research. If you're interested in the full paper, it's available here: https://lnkd.in/e7KvE68P