Decentralized Federated Learning with Blockchain for Privacy-Preserving Edge AI in Smart Cities

With all the new IoT devices and smart setups popping up in cities, we're generating a ton of data right at the network's edge. Now, while Edge AI can make decisions on the spot, training those AI models in a central way brings up issues around governance and privacy. So, this paper shares a pretty cool framework that mixes Federated Learning (FL) with Blockchain to let us train models on different devices in smart cities while keeping privacy intact. The idea with FL is that data stays put, and only the updates to the models get shared, which cuts down the risk of data leaks a lot. We also throw in a lightweight blockchain for keeping everything trustworthy and making sure the models are solid, even if some devices aren’t super reliable. This system not only scales well but also lower the amount of communication needed, all while keeping data secure and private. We used smart city data for our tests, and the results are promising better user privacy, tracking of data origins, and more accurate models, which all help in building secure and scalable Edge AI systems in our urban environments.