skip to main content

Title: “You Can’t Fix What You Can’t Measure”: Privately Measuring Demographic Performance Disparities in Federated Learning
Award ID(s):
1943584 1956435 2333448
Author(s) / Creator(s):
Date Published:
Journal Name:
Proceedings of Machine Learning Research
Page Range / eLocation ID:
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Natural gas distribution networks are part of a nation’s critical infrastructure, ensuring gas delivery to households and industries (e.g., power plants) with the correct chemical composition and the right conditions of pressure and temperature. Gas distribution is monitored and controlled by a Supervisory Control and Data Acquisition (SCADA) network, which provides centralized monitoring and control over the physical process.In this paper, we conduct the first openly available network measurement study of the SCADA network of an operational large-scale natural gas distribution network. With a total of 154 remote substations communicating through the SCADA system with a Control Room and over 98 days of observation, this is, to the best of our knowledge, the most extensive dataset of this kind analyzed to date.By combining the information obtained from engineering and IEC 104 network traffic, we reconstruct the gas distribution system’s layout, including the type and purpose of the substations and the physical properties of the gas that enters the SCADA system. Our analysis shows that it is possible to extract this information, essential for security monitoring, purely from the raw network traffic and without background knowledge provided by the control system engineers. We also note that configuration changes in SCADA environments, although probably less frequent than in IT environments, are not as rare and exceptional as the research community assumed. 
    more » « less
  2. In statistical mechanics, measuring the number of available states and their probabilities, and thus the system’s entropy, enables the prediction of the macroscopic properties of a physical system at equilibrium. This predictive capacity hinges on the knowledge of the a priori probabilities of observing the states of the system, given by the Boltzmann distribution. Unfortunately, the successes of equilibrium statistical mechanics are hardto replicate out of equilibrium, where the a priori probabilities of observing states are, in general, not known, precluding the naı̈ve application of common tools. In the last decade, exciting developments have occurred that enable direct numerical estimation of the entropy and density of states of athermal and non-equilibrium systems, thanks to significant methodological advances in the computation of the volume of high-dimensional basins of attraction. Here, we provide a detailed account of these methods, underscoring the challenges present in such estimations, recent progress on the matter, and promising directions for future work. 
    more » « less
  3. As we prepare new engineers to take on the fourth industrial revolution, engineering faculty are tasked with selecting, learning, evaluating, using, and teaching new technologies to apprentice engineers. To understand how these important tasks are being achieved, 21 engineering faculty members in a STEMfocused Midwest US university were interviewed. Engineering faculty showed an awareness of the rhetorical power of manuals and other instructional resources. Unfortunately, these resources are often inadequately designed to meet the unique needs of engineering faculty. In this paper, we propose that there is an exigency for a faculty-focused subgenre of instructional writing which addresses the needs of engineering instructors who teach students how to use technology while simultaneously learning how to use it themselves. Because of the overwhelming roles that faculty perform, we propose that the composition of this sub-genre should be the duty of technical writers who work closely with technology developers and engineering faculty. We forward that such a subgenre may find space in digital and non-digital learning resources through the inclusion of both the technical information necessary to use the technology, as well as pedagogical tools and activities to support student learning. These materials should be released in accordance with technology updates to ensure faculty are best positioned to teach the most current technologies. The proposed faculty-focused instructional writing subgenre may have implications beyond engineering education, because the need for learning resources may not be unique to engineering faculty, and likely exists for all university faculty learning and introducing new technologies within their courses. 
    more » « less