| Abstract |
In a coalition military setting, heterogeneous machine learning agents/models should be able to communicate with each other and share knowledge under information flow constraints to make informed accurate decisions. We present a method, wherein one deep neural network agent can share the information it gained by learning from a set of input data, with another deep neural network agent, which does not have access to the same input data. The second agent can use this shared knowledge to train itself and make decisions on similar data that it encounters on a mission, without having ever seen the data during its own training phase. We also develop an inter-layer similarity metric using which the agents can find the most similar layers between them. Information is shared between the models at these layers. We show that this framework presents a notion of machine transparency and operational machine-to-machine interpretability. |