Abstract: TinyML Devices like Arduino's and Raspberry Pi's are resource-constrained. This means that they are usually small and battery-powered and have low computation power and memory. Deploying modern neural network models on such devices is next to impossible due to how large they are both in terms of memory and the number of operations needs to execute them. This means that to deploy NNs on TinyML devices, we need to optimize them and scale them down.
There are many such algorithms, but the tools landscape is fragmented with different frameworks supporting different algorithms and only on their models. Moreover, they only support few algorithms and not the latest, better performing algorithms. Scaledown is attempting to bridge that gap and build a framework that helps you take models trained in any framework, optimize it using the latest algorithms, and then deploy it to TinyML devices.
To help you better understand this research field, we also do a lot of community work like hosting workshops and study groups, publishing free books and courses. We also help grow the field by doing research and adopting good Tiny MLOps practices in our framework.
Bio: Archana works as an AI Engineer at Continental Automotive. Her field of work is in TinyML i.e applying machine learning models to small devices with low power and memory requirements. This means that microcontrollers excite her and she loves working in this applied AI field. After work, you can usually find her volunteering at Women Who Code, where she co-leads the cloud and python track as a Leadership Fellow. Apart from that, she actively participates in TinyML and Women in Machine Learning events.
Artificial Intelligence Engineer | Leadership Fellow | Continental | Women Who Code