No abstract

No abstract

Today's general-purpose deep convolutional neural networks (CNN) for image classification and object detection are trained offline on large static datasets. Some applications, however, will require training in real-time on live video streams with a human-in-the-loop. We refer to this class of problem as Time-ordered Online Training (ToOT)-these problems will require a consideration of not only the quantity of incoming training data, but the human effort required to tag and use it. In this paper, we define training benefit as a metric to measure the effectiveness of a sequence in using each user interaction. We demonstrate and evaluate a system tailored to performing ToOT in the field, capable of training an image classifier on a live video stream through minimal input from a human operator. We show that by exploiting the time-ordered nature of the video stream through optical flow-based object tracking, we can increase the effectiveness of human actions by about 8 times.

A new economy built on the massive growth of endpoints on the internet will require precise and verifiable timing in ways that current systems do not support. Applications, computers, and communications systems have been developed with modules and layers that optimize data processing but degrade accurate timing. State-of-the-art systems now use timing only as a performance metric. Correctness of timing as a metric cannot currently be designed into systems independent of hardware and/or software implementations. To enable the massive growth predicted, accurate timing needs cross-disciplinary research to be integrated into these existing systems. This paper reviews the state of the art in six crucial areas central to the use of timing signals in these systems. Each area is shown to have critical issues requiring accuracy or integrity levels of timing, that need research contributions from a range of disciplines to solve.