More Like this

1. On the Impact of Random Seeds on the Fairness of Clinical Classifiers

   Amir, Silvio; van de Meent, Jan-Willem; Wallace, Byron C. (April 2021, North American Chapter of the Association for Computational Linguistics (NAACL))

   null (Ed.)

   Recent work has shown that fine-tuning large networks is surprisingly sensitive to changes in random seed(s). We explore the implications of this phenomenon for model fairness across demographic groups in clinical prediction tasks over electronic health records (EHR) in MIMIC-III —— the standard dataset in clinical NLP research. Apparent subgroup performance varies substantially for seeds that yield similar overall performance, although there is no evidence of a trade-off between overall and subgroup performance. However, we also find that the small sample sizes inherent to looking at intersections of minority groups and somewhat rare conditions limit our ability to accurately estimate disparities. Further, we find that jointly optimizing for high overall performance and low disparities does not yield statistically significant improvements. Our results suggest that fairness work using MIMIC-III should carefully account for variations in apparent differences that may arise from stochasticity and small sample sizes. 

   more » « less
2. The influence of palladium on the hydrogenation of acetylene on Ag(111)

   https://doi.org/10.1063/5.0050587  

   Molina, David L.; Muir, Mark; Abdel-Rahman, Mohammed K.; Trenary, Michael (May 2021, The Journal of Chemical Physics)
3. On the Impacts of Flow on the Migration and Growth of Cancer Cells

   https://doi.org/10.1115/dmd2022-1050  

   Akerkouch, Lahcen; Le, Trung; Jasuja, Haneesh; Katti, Kalpana; Katti, Dinesh (April 2022, Proceedings of the 2022 Design of Medical Devices Conference)

   Our study aims to identify the role of fluid flow in the growth of human bone cancer cells during metastasis. In our experiments, the cancer cells are seeded on the surface of cylindrical scaffolds in a bioreactor. The flow is laminar flow, which mimics the physiological conditions of the human body. A full-scale 3D high-resolution computational mesh of scaffold was created based on the physical scaffold's Micro-CT scans using open-source imaging software Slicer3D and Meshmixer. To investigate the influences of the flow on the seeded cells, we performed Computational Fluid Dynamics (CFD) simulations with the immersed boundary method (Gilmanov, Le, Sotiropoulos, JCP 300, 1, 2015). The computational domain was generated using the commercial software Gridgen. Our results show that the fluid flow velocity is highly dependent on the shape and pore sizes. In addition, the magnitude of the velocity on the surface where the cells are seeded is in between [0-0.05] μm/sallowing the cells to grow without being detached from the surface of the scaffold. Our future work will focus on (i) investigating the role of the shear stress on the distribution and orientation of the cancer cells. (ii) Simulating multiple scaffolds within the bioreactor to further quantify the impact of the gap on the flow velocity and shear. 

   more » « less
4. On the Impact of Relativistic Gravity on the Rate of Tidal Disruption Events

   https://doi.org/10.3847/1538-4357/ac85b3  

   Coughlin, Eric R.; Nixon, C. J. (August 2022, The Astrophysical Journal)

   Abstract The tidal disruption of stars by supermassive black holes (SMBHs) probes relativistic gravity. In the coming decade, the number of observed tidal disruption events (TDEs) will grow by several orders of magnitude, allowing statistical inferences of the properties of the SMBH and stellar populations. Here we analyze the probability distribution functions of the pericenter distances of stars that encounter an SMBH in the Schwarzschild geometry, where the results are completely analytic, and the Kerr metric. From this analysis we calculate the number of observable TDEs, defined to be those that come within the tidal radiusr_tbut outside the direct capture radius (which is, in general, larger than the horizon radius). We find that relativistic effects result in a steep decline in the number of stars that have pericenter distancesr_p≲ 10r_g, wherer_g=GM/c², and that for maximally spinning SMBHs the distribution function ofr_pat such distances scales as $f_{{\mathrm{r}}_{\mathrm{p}}}\propto r_{\mathrm{p}}^{4/3}$, or in terms ofβ≡r_t/r_pscales asf_β∝β^−10/3. We find that spin has little effect on the TDE fraction until the very-high-mass end, where instead of being identically zero the rate is small (≲1% of the expected rate in the absence of relativistic effects). Effectively independent of spin, if the progenitors of TDEs reflect the predominantly low-mass stellar population and thus have masses ≲1M_⊙, we expect a substantial reduction in the rate of TDEs above 10⁷M_⊙. 

   more » « less
5. On the Local Linear Rate of Consensus on the Stiefel Manifold

   https://doi.org/10.1109/TAC.2023.3330735  

   Chen, Shixiang; Garcia, Alfredo; Hong, Mingyi; Shahrampour, Shahin (November 2023, IEEE Transactions on Automatic Control)