Abstract
An Internet-of-Things (IoT) platform that enables the retraining of machine learning models on embedded devices is described. The IoT platform utilizes transfer learning to retrain models in a cluster of IoT products connected to each-other in a local-area network (LAN), personal-area network (PAN), or wireless personal-area network (WPAN), to be reused for a similar purpose. Unlike current IoT platforms, the distributed transfer learning IoT platform does not need to utilize a centralized computing system, such as a cloud-computing server or a network server to perform model training, but rather execute this training in the cluster of IoT products. To reach this goal, in addition to transfer learning, the described IoT platform supports application programming interfaces (APIs) that specify a small portion of the existing pretrained model to be retrained, specify a data pipeline in the cluster of IoT devices to be used to retrain the model, and tune the model.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Singh, Hanumant Prasad, "Distributed Transfer Learning on Embedded Devices", Technical Disclosure Commons, (November 28, 2018)
https://www.tdcommons.org/dpubs_series/1709