Tiny AI

Tiny AI (or TinyML) miniaturizes powerful machine learning models, enabling real-time, energy-efficient inference directly on resource-constrained edge devices: think microcontrollers and wearables.

Tiny AI is the strategic answer to the 'Big AI' problem: high energy use, privacy concerns, and cloud reliance. This technology focuses on model compression techniques (pruning, quantization, and knowledge distillation) to shrink algorithms by factors up to 10x, allowing them to run on hardware with less than one milliwatt of power. For example, Huawei’s TinyBERT achieved 96% of the original BERT model's performance while being 7.5 times smaller. The shift to local processing enhances data privacy and supports critical applications like connected health solutions and Industry 4.0 cobots, driving sustainable, decentralized intelligence at the extreme edge.

https://www.imec-int.com/en/tiny-ai