Cooperative Learning with Visual Attributes
This addresses the problem of enabling machine learning in scenarios with privacy, security, or bandwidth constraints by allowing agents to share knowledge without exchanging raw data.
The paper tackles the problem of learning visual concepts with supervision from a peer machine instead of a human teacher, using visual attributes as a communication mode, and shows that cooperative learning improves performance for both agents compared to isolated learning across objects, scenes, and animals domains.
Learning paradigms involving varying levels of supervision have received a lot of interest within the computer vision and machine learning communities. The supervisory information is typically considered to come from a human supervisor -- a "teacher" figure. In this paper, we consider an alternate source of supervision -- a "peer" -- i.e. a different machine. We introduce cooperative learning, where two agents trying to learn the same visual concepts, but in potentially different environments using different sources of data (sensors), communicate their current knowledge of these concepts to each other. Given the distinct sources of data in both agents, the mode of communication between the two agents is not obvious. We propose the use of visual attributes -- semantic mid-level visual properties such as furry, wooden, etc.-- as the mode of communication between the agents. Our experiments in three domains -- objects, scenes, and animals -- demonstrate that our proposed cooperative learning approach improves the performance of both agents as compared to their performance if they were to learn in isolation. Our approach is particularly applicable in scenarios where privacy, security and/or bandwidth constraints restrict the amount and type of information the two agents can exchange.