skip to main content


Search for: All records

Creators/Authors contains: "Wu, Kun"

Note: When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external site maintained by the publisher. Some full text articles may not yet be available without a charge during the embargo (administrative interval).
What is a DOI Number?

Some links on this page may take you to non-federal websites. Their policies may differ from this site.

  1. Graph Neural Networks (GNNs) have been widely used in various graph-based applications. Recent studies have shown that GNNs are vulnerable to link-level membership inference attacks (LMIA) which can infer whether a given link was included in the training graph of a GNN model. While most of the studies focus on the privacy vulnerability of the links in the entire graph, none have inspected the privacy risk of specific subgroups of links (e.g., links between LGBT users). In this paper, we present the first study of disparity in subgroup vulnerability (DSV) of GNNs against LMIA. First, with extensive empirical evaluation, we demonstrate the existence of non-negligible DSV under various settings of GNN models and input graphs. Second, by both statistical and causal analysis, we identify the difference between three specific graph structural properties of subgroups as one of the underlying reasons for DSV. Among the three properties, the difference between subgroup density has the largest causal effect on DSV. Third, inspired by the causal analysis, we design a new defense mechanism named FairDefense to mitigate DSV while providing protection against LMIA. At a high level, at each iteration of target model training, FairDefense randomizes the membership of edges in the training graph with a given probability, aiming to reduce the gap between the density of different subgroups for DSV mitigation. Our empirical results demonstrate that FairDefense outperforms the existing defense methods in the trade-off between defense and target model accuracy. More importantly, it offers better DSV mitigation.

     
    more » « less
    Free, publicly-accessible full text available October 1, 2024
  2. Free, publicly-accessible full text available August 4, 2024
  3. Abstract

    This study investigates the global distribution of electron temperature enhancement observed by Defense Meteorological Satellite Program F16 satellite and its dependence on the season and solar activity for the solar maximum (2014) and minimum (2018) years during geomagnetic quiet times (maximum per day ap <10). Electron temperature enhancements occurred mainly over the North American‐Atlantic (260°–360°E) and Eurasia (0°–160°E) (Southern Oceania (80°–280°E)) sector in the Northern (Southern) Hemisphere and are prominent in the winter hemispheres and solar maximum year. They have obvious longitude characteristics. Interestingly, they could extend to geomagnetic equatorial regions in the North American‐Atlantic sector from high to low latitudes in the December Solstice, further crossed the magnetic equator, and merged into the Southern Hemisphere in 2014, where the maximum temperature reached ∼3500 K. Our analysis indicates that low‐energy electrons (<100 eV) associated with photoelectron from the conjugate sunlit hemisphere, can contribute to these enhancements. Furthermore, the local geomagnetic declination, magnetic equator position, and terminator position at magnetic conjugate points together can impact the global distribution of photoelectrons of different energies and therefore the electron temperature enhancement distribution. Other processes (including local electron density variation) may play certain roles as well.

     
    more » « less
    Free, publicly-accessible full text available September 1, 2024
  4. Active fluids have potential applications in micromixing, but little is known about the mixing kinematics of such systems with spatiotemporally-varying activity. To investigate, UV-activated caged ATP was used to activate controlled regions of microtubule-kinesin active fluid inducing a propagating active-passive interface. The mixing process of the system from non-uniform to uniform activity as the interface advanced was observed with fluorescent tracers and molecular dyes. At low Péclet numbers (diffusive transport), the active-inactive interface progressed toward the inactive area in a diffusion-like manner and at high Péclet numbers (convective transport), the active-inactive interface progressed in a superdiffusion-like manner. The results show mixing in non-uniform active fluid systems evolve from a complex interplay between the spatial distribution of ATP and its active transport. This active transport may be diffusion-like or superdiffusion-like depending on Péclet number and couples the spatiotemporal distribution of ATP and the subsequent localized active stresses of active fluid. Our work will inform the design of future microfluidic mixing applications and provide insight into intracellular mixing processes. *T.E.B., E.H.T., J.H.D., and K.-T.W. acknowledge support from the National Science Foundation (NSF-CBET-2045621). C.-C. C. was supported through the National Science and Technology Council (NSTC), Taiwan (111-2221-E-006-102-MY3). M.M.N. was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences (DE-SC0022280). 
    more » « less
  5. Active fluids with spatiotemporally varying activity have potential applications to micromixing; however previously existing active fluids models are not prepared to account for spatiotemporally-varying active stresses. Our experimental work used UV-activated caged ATP to activate controlled regions of microtubule-kinesin active fluid inducing a propagating active-passive interface. Here, we recapitulate our experimental results with two models. The first model redistributes an initial ATP distribution by Fick's law and translates the ATP distribution into a velocity profile by Michaelis-Menton kinetics. This model reproduces our experimental measurements for the low-Péclet number limit within 10% error without fitting parameters. However, as the model is diffusion based, it fails to capture the convective based superdiffusive-like behaviour at high Péclet numbers. Our second model introduces a spatiotemporally varying ATP field to an existing nematohydrodynamic active fluid model and then couples the active stresses to local ATP concentrations. This model is successful in qualitatively capturing the superdiffusive-like progression of the active-inactive interface for high Peclet number (convective transport) experimental cases. Our results show that new model frameworks are necessary for capturing the behaviour of active fluid with spatiotemporally varying activity. *T.E.B., E.H.T., J.H.D., and K.-T.W. acknowledge support from the National Science Foundation (NSF-CBET-2045621). C.-C. C. was supported through the National Science and Technology Council (NSTC), Taiwan (111-2221-E-006-102-MY3). M.M.N. was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences (DE-SC0022280). 
    more » « less
  6. Abstract Active fluids have applications in micromixing, but little is known about the mixing kinematics of systems with spatiotemporally-varying activity. To investigate, UV-activated caged ATP is used to activate controlled regions of microtubule-kinesin active fluid and the mixing process is observed with fluorescent tracers and molecular dyes. At low Péclet numbers (diffusive transport), the active-inactive interface progresses toward the inactive area in a diffusion-like manner that is described by a simple model combining diffusion with Michaelis-Menten kinetics. At high Péclet numbers (convective transport), the active-inactive interface progresses in a superdiffusion-like manner that is qualitatively captured by an active-fluid hydrodynamic model coupled to ATP transport. Results show that active fluid mixing involves complex coupling between distribution of active stress and active transport of ATP and reduces mixing time for suspended components with decreased impact of initial component distribution. This work will inform application of active fluids to promote micromixing in microfluidic devices. 
    more » « less
  7. Fluid mixing is driven by the passive process of diffusion and the active process of stretching and folding, which homogenize the system's constituents. Conventionally, the active process is applied via external shearing machines such as a kitchen stand mixer. However, applying external shearing becomes more challenging in mesoscopic fluid systems due to the increasing difficulty of controlling the injection of energy on the micron scale. To overcome this challenge, we introduced microtubule-kinesin active fluid to power the active mixing process. To demonstrate its mixing capability, we created a multi-fluid system where active fluid is adjacent to an inactivated, passive fluid and allowed the active fluid to blend with the passive fluid until the system reaches a homogeneous state. We found that the mixing dynamics of such active-passive fluid mixing was dominated by the passive process of diffusion, until the activity of active fluid was tuned to be sufficiently high and the active processes of active fluid began to dominate the mixing process. Our work will stimulate the development of utilizing active fluid to accomplish mesoscale mixing tasks in multi-fluid systems at the micron scale. *We acknowledge support from the National Science Foundation (NSF-CBET-2045621). 
    more » « less