Machine Learning for Network-Denied Environments

Cloud-based AI/ML solutions are powerful and up-to-date but rely on network availability to operate. This is problematic to achieve in places where these services would be greatly beneficial such as in conflict zones, major disaster areas, and under-developed regions. Meanwhile, client-based AI/ML solutions (e.g. MobileNet) can run without network connection but cannot be updated very easily in such environments. Machine learning In Network-Denied Environments (MINDE) is a project that aims to showcase the feasibility of a hybrid approach combining a cloud-based server and web client with a mobile client capable of working offline. In particular, MINDE focuses on the problem of automatically classifying (i.e., labeling) the central object in pictures. The mobile client enables the collecting and classification of images directly from the phone’s camera or from its photo library. If the automatic classification is not correct, the user is able to relabel it, potentially with a label that has never been seen before. When a network connection is available, the mobile client will upload new images to the server along with any user-supplied labels. The server and its web client allow users to retrain/fine-tune the classification model with this new data and send the new model back to the mobile client. Alternatively, when communication with the server is not possible, the mobile client will attempt to use peer-to-peer communication with another mobile client to exchange images, labels and updated models.

Student Team
  • Sanjog Baniya
  • Jonathon M Dooley
  • Enrico Efendi
  • Wilson Gan
  • Xavier Lara
  • Kevin Maravillas
  • Howard Nguyen
  • Nisapat Poolkwan
  • Johnson Tan
  • Justin To
  • Alvin Yu
Project Sponsor
Project Liaisons
  • Mark Core
  • Benjamin Nye
Faculty Advisors
  • Chengyu Sun