Search

There are demographic biases present in current facial recognition (FR) models. To measure these biases across different ethnic and gender subgroups, we introduce our Balanced Faces in the Wild (BFW) dataset. This dataset allows for the characterization of FR performance per subgroup. We found that relying on a single score threshold to differentiate between genuine and imposters sample pairs leads to suboptimal results. Additionally, performance within subgroups often varies significantly from the global average. Therefore, specific error rates only hold for populations that match the validation data. To mitigate imbalanced performances, we propose a novel domain adaptation learning scheme that uses facial features extracted from state-of-the-art neural networks. This scheme boosts the average performance and preserves identity information while removing demographic knowledge. Removing demographic knowledge prevents potential biases from affecting decision-making and protects privacy by eliminating demographic information. We explore the proposed method and demonstrate that subgroup classifiers can no longer learn from features projected using our domain adaptation scheme. For access to the source code and data, please visit https://github.com/visionjo/facerec-bias-bfw., Comment: arXiv admin note: text overlap with arXiv:2102.08941

It is widely recognized that the shape of epithelial cells is determined by the tension generated by the actomyosin cortex and the adhesion of cells to the substrate and to each other. To account for these biological and structural contributions to cell shape, different physical models have been proposed. However, an experimental procedure that would allow a validation of a minimal physical model for the shape of epithelial cells in 3D has not yet been proposed. In this study, we cultured MDCK epithelial cells on substrates with a sinusoidal profile, allowing us to measure the shape of the cells on various positive and negative curvatures. We found that MDCK cells are thicker in the valleys than on the crests of sinusoidal substrates. The influence of curvature on the shape of epithelial cells could not be understood with a model using only differential apical, basal and lateral surface energies. However, the addition of an apical line tension was sufficient to quantitatively account for the experimental measurements. The model also accounts for the shape of MDCK cells that overexpress E-cadherin. On the other hand, when reducing myosin II activity with blebbistatin, we measured a saturation of the difference in cell thickness between valleys and crests, suggesting the need for a term limiting large cell deformations. Our results show that a minimal model that accounts for epithelial cell shape needs to include an apical line tension in addition to differential surface energies, highlighting the importance of structures that produce anisotropic tension in epithelial cells, such as the actin belt linking adherens junctions. In the future, our experimental procedure could be used to test a wider range of physical models for the shape of epithelia in curved environments, including, for example, continuous models., Comment: 15 pages, 10 figures

We reveal critical insights into problems of bias in state-of-the-art facial recognition (FR) systems using a novel Balanced Faces In the Wild (BFW) dataset: data balanced for gender and ethnic groups. We show variations in the optimal scoring threshold for face-pairs across different subgroups. Thus, the conventional approach of learning a global threshold for all pairs resulting in performance gaps among subgroups. By learning subgroup-specific thresholds, we not only mitigate problems in performance gaps but also show a notable boost in the overall performance. Furthermore, we do a human evaluation to measure the bias in humans, which supports the hypothesis that such a bias exists in human perception. For the BFW database, source code, and more, visit github.com/visionjo/facerec-bias-bfw., Comment: Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2020

In this paper we present a massively parallel open source solver for Richards equation, named the RichardsFOAM solver. This solver has been developed in the framework of the open source generalist computational fluid dynamics tool box OpenFOAM (R) and is capable to deal with large scale problems in both space and time. The source code for RichardsFOAM may be downloaded from the CPC program library website.It exhibits good parallel performances (up to $\sim$90% parallel efficiency with 1024 processors both in strong and weak scaling), and the conditions required for obtaining such performances are analysed and discussed. These performances enable the mechanistic modelling of water fluxes at the scale of experimental watersheds (up to few square kilometres of surface area), and on time scales of decades to a century. Such a solver can be useful in various applications, such as environmental engineering for long term transport of pollutants in soils, water engineering for assessing the impact of land settlement on water resources, or in the study of weathering processes on the watersheds.

Comprehensive Two dimensional gas chromatography (GCxGC) plays a central role into the elucidation of complex samples. The automation of the identification of peak areas is of prime interest to obtain a fast and repeatable analysis of chromatograms. To determine the concentration of compounds or pseudo-compounds, templates of blobs are defined and superimposed on a reference chromatogram. The templates then need to be modified when different chromatograms are recorded. In this study, we present a chromatogram and template alignment method based on peak registration called BARCHAN. Peaks are identified using a robust mathematical morphology tool. The alignment is performed by a probabilistic estimation of a rigid transformation along the first dimension, and a non-rigid transformation in the second dimension, taking into account noise, outliers and missing peaks in a fully automated way. Resulting aligned chromatograms and masks are presented on two datasets. The proposed algorithm proves to be fast and reliable. It significantly reduces the time to results for GCxGC analysis., Comment: 15 pages, published in the Special issue for RIVA 2016, 40th International Symposium on Capillary Chromatography and 13th GCxGC Symposium

This paper is an English translation of Michel H\'enon's thesis, "Sur l'\'evolution dynamique des amas globulaires" originally published in French in the Annales d'Astrophysique, Vol. 24, p.369 (1961)., Comment: Originally published in 1961. 44 pages, 23 figures

This paper is an English translation of Michel H\'enon's article, "Sur l'\'evolution dynamique des amas globulaires. II- L'amas isol\'e" originally published in French in the Annales d'Astrophysique, Vol. 28, p.62 (1965). A translation of the first paper of this series ("Sur l'\'evolution dynamique des amas globulaires", H\'enon 1961, Annales d'Astrophysique, Vol. 24, p.369) is also available., Comment: Originally published in 1965. 7 pages, 5 figures

We compare and synthesize the results of two microrheological experiments on the cytoskeleton of single cells. In the first one, the creep function J(t) of a cell stretched between two glass plates is measured after applying a constant force step. In the second one, a micrometric bead specifically bound to transmembrane receptors is driven by an oscillating optical trap, and the viscoelastic coefficient $G_e(\omega)$ is retrieved. Both $J(t)$ and $G_e(\omega)$ exhibit power law behavior: $J(t)= A(t/t_0)^\alpha$ and $\bar G_e(\omega)\bar = G_0 (\omega/\omega_0)^\alpha$, with the same exponent $\alpha\approx 0.2$. This power law behavior is very robust ; $\alpha$ is distributed over a narrow range, and shows almost no dependance on the cell type, on the nature of the protein complex which transmits the mechanical stress, nor on the typical length scale of the experiment. On the contrary, the prefactors $A_0$ and $G_0$appear very sensitive to these parameters. Whereas the exponents $\alpha$ are normally distributed over the cell population, the prefactors $A_0$ and $G_0$ follow a log-normal repartition. These results are compared with other data published in the litterature. We propose a global interpretation, based on a semi-phenomenological model, which involves a broad distribution of relaxation times in the system. The model predicts the power law behavior and the statistical repartition of the mechanical parameters, as experimentally observed for the cells. Moreover, it leads to an estimate of the largest response time in the cytoskeletal network: $\tau_m \approx 1000$ s., Comment: 47 pages, 14 figures // v2: PDF file is now Acrobat Reader 4 (and up) compatible // v3: Minor typos corrected - The presentation of the model have been substantially rewritten (p. 17-18), in order to give more details - Enhanced description of protocols // v4: Minor corrections in the text : the immersion angles are estimated and not measured // v5: Minor typos corrected. Two references were clarified

We show that the deterministic past history of the Universe can be uniquely reconstructed from the knowledge of the present mass density field, the latter being inferred from the 3D distribution of luminous matter, assumed to be tracing the distribution of dark matter up to a known bias. Reconstruction ceases to be unique below those scales -- a few Mpc -- where multi-streaming becomes significant. Above 6 Mpc/h we propose and implement an effective Monge-Ampere-Kantorovich method of unique reconstruction. At such scales the Zel'dovich approximation is well satisfied and reconstruction becomes an instance of optimal mass transportation, a problem which goes back to Monge (1781). After discretization into N point masses one obtains an assignment problem that can be handled by effective algorithms with not more than cubic time complexity in N and reasonable CPU time requirements. Testing against N-body cosmological simulations gives over 60% of exactly reconstructed points. We apply several interrelated tools from optimization theory that were not used in cosmological reconstruction before, such as the Monge-Ampere equation, its relation to the mass transportation problem, the Kantorovich duality and the auction algorithm for optimal assignment. Self-contained discussion of relevant notions and techniques is provided., Comment: 26 pages, 14 figures; accepted to MNRAS. Version 2: numerous minour clarifications in the text, additional material on the history of the Monge-Ampere equation, improved description of the auction algorithm, updated bibliography. Version 3: several misprints corrected

We describe a mechanical device which can be used as an analog computer to solve the transportation problem. In practice this device is simulated by a numerical algorithm. Tests show that this algorithm is 60 times faster than a current subroutine (NAG library) for an average 1000 x 1000 problem. Its performance is even better for degenerate problems in which the weights take only a small number of integer values., Comment: 45 pages, 10 figures, 4 tables. This paper was written in 1992, but not published (see Addendum for a brief history)

Numerical integrations in celestial mechanics often involve the repeated computation of a rotation with a constant angle. A direct evaluation of these rotations yields a linear drift of the distance to the origin. This is due to roundoff in the representation of the sine s and cosine c of the angle theta. In a computer, one generally gets c^2 + s^2 <> 1, resulting in a mapping that is slightly contracting or expanding. In the present paper we present a method to find pairs of representable real numbers s and c such that c^2 + s^2 is as close to 1 as possible. We show that this results in a drastic decrease of the systematic error, making it negligible compared to the random error of other operations. We also verify that this approach gives good results in a realistic celestial mechanics integration., Comment: 24 pages, 3 figures