-
Demographic Bias of Expert-Level Vision-Language Foundation Models in Medical Imaging
Authors:
Yuzhe Yang,
Yujia Liu,
Xin Liu,
Avanti Gulhane,
Domenico Mastrodicasa,
Wei Wu,
Edward J Wang,
Dushyant W Sahani,
Shwetak Patel
Abstract:
Advances in artificial intelligence (AI) have achieved expert-level performance in medical imaging applications. Notably, self-supervised vision-language foundation models can detect a broad spectrum of pathologies without relying on explicit training annotations. However, it is crucial to ensure that these AI models do not mirror or amplify human biases, thereby disadvantaging historically margin…
▽ More
Advances in artificial intelligence (AI) have achieved expert-level performance in medical imaging applications. Notably, self-supervised vision-language foundation models can detect a broad spectrum of pathologies without relying on explicit training annotations. However, it is crucial to ensure that these AI models do not mirror or amplify human biases, thereby disadvantaging historically marginalized groups such as females or Black patients. The manifestation of such biases could systematically delay essential medical care for certain patient subgroups. In this study, we investigate the algorithmic fairness of state-of-the-art vision-language foundation models in chest X-ray diagnosis across five globally-sourced datasets. Our findings reveal that compared to board-certified radiologists, these foundation models consistently underdiagnose marginalized groups, with even higher rates seen in intersectional subgroups, such as Black female patients. Such demographic biases present over a wide range of pathologies and demographic attributes. Further analysis of the model embedding uncovers its significant encoding of demographic information. Deploying AI systems with these biases in medical imaging can intensify pre-existing care disparities, posing potential challenges to equitable healthcare access and raising ethical questions about their clinical application.
△ Less
Submitted 22 February, 2024;
originally announced February 2024.
-
Observation of time-reversal symmetric Hall effect in graphene-WSe2 heterostructures at room temperature
Authors:
Priya Tiwari,
Divya Sahani,
Atasi Chakraborty,
Kamal Das,
Kenji Watanabe,
Takashi Taniguchi,
Amit Agarwal,
Aveek Bid
Abstract:
In this letter, we provide experimental evidence of the time-reversal symmetric Hall effect in a mesoscopic system, namely high-mobility graphene/WSe$_2$ heterostructures. This linear, dissipative Hall effect, whose sign depends on the sign of the charge carriers, persists up to room temperature. The magnitude and the sign of the Hall signal can be tuned using an external perpendicular electric fi…
▽ More
In this letter, we provide experimental evidence of the time-reversal symmetric Hall effect in a mesoscopic system, namely high-mobility graphene/WSe$_2$ heterostructures. This linear, dissipative Hall effect, whose sign depends on the sign of the charge carriers, persists up to room temperature. The magnitude and the sign of the Hall signal can be tuned using an external perpendicular electric field. Our joint experimental and theoretical study establishes that the strain induced by lattice mismatch, or angle inhomogeneity, produces anisotropic bands in graphene while simultaneously breaking the inversion symmetry. The band anisotropy and reduced spatial symmetry lead to the appearance of a time-reversal symmetric Hall effect. Our study establishes graphene-transition metal dichalcogenide-based heterostructures as an excellent platform for studying the effects of broken symmetry on the physical properties of band-engineered two-dimensional systems.
△ Less
Submitted 18 July, 2023; v1 submitted 5 January, 2023;
originally announced January 2023.
-
Transition from three- to two-dimensional Ising superconductivity in few-layer NbSe2 by proximity effect from van der Waals heterostacking
Authors:
Prakiran Baidya,
Divya Sahani,
Hemanta Kumar Kundu,
Simrandeep Kaur,
Priya Tiwari,
Vivas Bagwe,
John Jesudasan,
Awadhesh Narayan,
Pratap Raychaudhuri,
Aveek Bid
Abstract:
We report the experimental observation of Ising superconductivity in 3-dimensional NbSe2 stacked with single-layer MoS2. The angular dependence of the upper critical magnetic field and the temperature dependence of the upper parallel critical field confirm the appearance of two-dimensional Ising superconductivity in the 3-dimensional NbSe2 with single-layer MoS2 overlay. We show that the supercond…
▽ More
We report the experimental observation of Ising superconductivity in 3-dimensional NbSe2 stacked with single-layer MoS2. The angular dependence of the upper critical magnetic field and the temperature dependence of the upper parallel critical field confirm the appearance of two-dimensional Ising superconductivity in the 3-dimensional NbSe2 with single-layer MoS2 overlay. We show that the superconducting phase has strong Ising spin-orbit correlations which make the holes spin non-degenerate. Our observation of Ising superconductivity in heterostructures of few-layer NbSe2 of thickness ~ 15 nm with single-layer MoS2 raises the interesting prospect of observing topological chiral superconductors with nontrivial Chern numbers in a momentum-space spin-split fermionic system.
△ Less
Submitted 12 November, 2021;
originally announced November 2021.
-
How many interchanges does the selection sort make for iid geometric(p) input?
Authors:
Debasish Sahani,
Soubhik Chakraborty
Abstract:
The note derives an expression for the number of interchanges made by selection sort when the sorting elements are iid variates from geometric distribution. Empirical results reveal we can work with a simpler model compared to what is suggestive in theory. The morale is that statistical analysis of an algorithm's complexity has something to offer in its own right and should be therefore ventured n…
▽ More
The note derives an expression for the number of interchanges made by selection sort when the sorting elements are iid variates from geometric distribution. Empirical results reveal we can work with a simpler model compared to what is suggestive in theory. The morale is that statistical analysis of an algorithm's complexity has something to offer in its own right and should be therefore ventured not with a predetermined mindset to verify what we already know in theory. Herein also lies the concept of an empirical O, a novel although subjective bound estimate over a finite input range obtained by running computer experiments. For an arbitrary algorithm, where theoretical results could be tedious, this could be of greater use.
△ Less
Submitted 27 February, 2012;
originally announced February 2012.
