Search

Despite the superior performance of Deep Learning (DL) on numerous segmentation tasks, the DL-based approaches are notoriously overconfident about their prediction with highly polarized label probability. This is often not desirable for many applications with the inherent label ambiguity even in human annotations. This challenge has been addressed by leveraging multiple annotations per image and the segmentation uncertainty. However, multiple per-image annotations are often not available in a real-world application and the uncertainty does not provide full control on segmentation results to users. In this paper, we propose novel methods to improve the segmentation probability estimation without sacrificing performance in a real-world scenario that we have only one ambiguous annotation per image. We marginalize the estimated segmentation probability maps of networks that are encouraged to under-/over-segment with the varying Tversky loss without penalizing balanced segmentation. Moreover, we propose a unified hypernetwork ensemble method to alleviate the computational burden of training multiple networks. Our approaches successfully estimated the segmentation probability maps that reflected the underlying structures and provided the intuitive control on segmentation for the challenging 3D medical image segmentation. Although the main focus of our proposed methods is not to improve the binary segmentation performance, our approaches marginally outperformed the state-of-the-arts. The codes are available at \url{https://github.com/sh4174/HypernetEnsemble}.

Recurrent neural networks (RNNs) are a popular choice for modeling sequential data. Modern RNN architectures assume constant time-intervals between observations. However, in many datasets (e.g. medical records) observation times are irregular and can carry important information. To address this challenge, we propose continuous recurrent units (CRUs) -- a neural architecture that can naturally handle irregular intervals between observations. The CRU assumes a hidden state, which evolves according to a linear stochastic differential equation and is integrated into an encoder-decoder framework. The recursive computations of the CRU can be derived using the continuous-discrete Kalman filter and are in closed form. The resulting recurrent architecture has temporal continuity between hidden states and a gating mechanism that can optimally integrate noisy observations. We derive an efficient parameterization scheme for the CRU that leads to a fast implementation f-CRU. We empirically study the CRU on a number of challenging datasets and find that it can interpolate irregular time series better than methods based on neural ordinary differential equations., Comment: Accepted at ICML 2022, Baltimore, Maryland

A multisection of a 4-manifold is a decomposition into 1-handlebodies intersecting pairwise along 3-dimensional handlebodies or along a central closed surface; this generalizes the Gay-Kirby trisections. We show how to compute the twisted absolute and relative homology, the torsion and the twisted intersection form of a 4-manifold from a multisection diagram. The homology and torsion are given by a complex of free modules defined by the diagram and the intersection form is expressed in terms of the intersection form on the central surface. We give efficient proofs, with very few computations, thanks to a retraction of the (possibly punctured) 4-manifold onto a CW-complex determined by the multisection diagram. Further, a multisection induces an open book decomposition on the boundary of the 4-manifold; we describe the action of the monodromy on the homology of the page from the multisection diagram., Comment: 17 pages, 6 figures, comments welcome

We present a new infrared survey covering the three Euclid deep fields and four other Euclid calibration fields using Spitzer's Infrared Array Camera (IRAC). We have combined these new observations with all relevant IRAC archival data of these fields in order to produce the deepest possible mosaics of these regions. In total, these observations represent nearly 11% of the total Spitzer mission time. The resulting mosaics cover a total of approximately 71.5deg$^2$ in the 3.6 and 4.5um bands, and approximately 21.8deg$^2$ in the 5.8 and 8um bands. They reach at least 24 AB magnitude (measured to sigma, in a 2.5 arcsec aperture) in the 3.6um band and up to ~ 5 mag deeper in the deepest regions. The astrometry is tied to the Gaia astrometric reference system, and the typical astrometric uncertainty for sources with 16<[3.6]<19 is <0.15 arcsec. The photometric calibration is in excellent agreement with previous WISE measurements. We have extracted source number counts from the 3.6um band mosaics and they are in excellent agreement with previous measurements. Given that the Spitzer Space Telescope has now been decommissioned these mosaics are likely to be the definitive reduction of these IRAC data. This survey therefore represents an essential first step in assembling multi-wavelength data on the Euclid deep fields which are set to become some of the premier fields for extragalactic astronomy in the 2020s., Comment: 15 pages with 11 figures, approved by Euclid Consortium Publication Board and submitted to Astronomy and Astrophysics. Data products will become available via the IRSA website once the paper is accepted. This paper is a companion to "COSMOS2020: A panchromatic view of the Universe to z~10 from two complementary catalogs" by John Weaver et al., which is being posted in parallel

Runtime monitoring is generally considered a light-weight alternative to formal verification. In safety-critical systems, however, the monitor itself is a critical component. For example, if the monitor is responsible for initiating emergency protocols, as proposed in a recent aviation standard, then the safety of the entire system critically depends on guarantees of the correctness of the monitor. In this paper, we present a verification extension to the Lola monitoring language that integrates the efficient specification of the monitor with Hoare-style annotations that guarantee the correctness of the monitor specification. We add two new operators, assume and assert, which specify assumptions of the monitor and expectations on its output, respectively. The validity of the annotations is established by an integrated SMT solver. We report on experience in applying the approach to specifications from the avionics domain, where the annotation with assumptions and assertions has lead to the discovery of safety-critical errors in the specifications. The errors range from incorrect default values in offset computations to complex algorithmic errors that result in unexpected temporal patterns.

We investigate the importance of lensing magnification for estimates of galaxy clustering and its cross-correlation with shear for the photometric sample of Euclid. Using updated specifications, we study the impact of lensing magnification on the constraints and the shift in the estimation of the best fitting cosmological parameters that we expect if this effect is neglected. We follow the prescriptions of the official Euclid Fisher matrix forecast for the photometric galaxy clustering analysis and the combination of photometric clustering and cosmic shear. The slope of the luminosity function (local count slope), which regulates the amplitude of the lensing magnification, and the galaxy bias have been estimated from the Euclid Flagship simulation.We find that magnification significantly affects both the best-fit estimation of cosmological parameters and the constraints in the galaxy clustering analysis of the photometric sample. In particular, including magnification in the analysis reduces the 1$\sigma$ errors on $\Omega_{\text{m},0}, w_{0}, w_a$ at the level of 20-35%, depending on how well we will be able to independently measure the local count slope. In addition, we find that neglecting magnification in the clustering analysis leads to shifts of up to 1.6$\sigma$ in the best-fit parameters. In the joint analysis of galaxy clustering, cosmic shear, and galaxy-galaxy lensing, magnification does not improve precision, but it leads to an up to 6$\sigma$ bias if neglected. Therefore, for all models considered in this work, magnification has to be included in the analysis of galaxy clustering and its cross-correlation with the shear signal ($3\times2$pt analysis) for an accurate parameter estimation., Comment: 26 pages, 11 figures. Version accepted for publication in A&A

We study the problem of robust performance of quantum systems under structured uncertainties. A specific feature of closed (Hamiltonian) quantum systems is that their poles lie on the imaginary axis and that neither a coherent controller nor physically relevant structured uncertainties can alter this situation. This changes for open systems where decoherence ensures asymptotic stability and creates a unique landscape of pure performance robustness, with the distinctive feature that closed-loop stability is secured by the underlying physics and needs not be enforced. This stability, however, is often detrimental to quantum-enhanced performance, and additive perturbations of the Hamiltonian give rise to dynamic generators that are nonlinear in the perturbed parameters, invalidating classical paradigms to assess robustness to structured perturbations such as singular value analysis. This problem is addressed using a fixed-point iteration approach to determine a maximum perturbation strength $\delta_{\max}$ that ensures that the transfer function remains bounded, $||T_\delta||<\delta^{-1}$ for $\delta<\delta_{\max}$., Comment: Accepted for IEEE CDC 2021

Spin-based metal-oxide-semiconductor field-effect transistors (MOSFET) with a high-mobility III-V channel are studied using self-consistent quantum corrected ensemble Monte Carlo device simulations of charge and spin transport. The simulations including spin-orbit coupling mechanisms (Dresselhaus and Rashba coupling) examine the electron spin transport in the 25 nm gate length In$_{0.7}$Ga$_{0.3}$As MOSFET. The transistor lateral dimensions (the gate length, the source-to-gate, and the gate-to-drain spacers) are increased to investigate the spin-dependent drain current modulation induced by the gate from room temperature of 300 K down to 77 K. This modulation increases with increasing temperature due to increased Rashba coupling. Finally, an increase of up to 20 nm in the gate length, source-to-gate, or the gate-to-drain spacers increases the spin polarization and enhances the spin-dependent drain current modulation at the drain due to polarization-refocusing effects.

We applied machine learning methods to predict chemical hazards focusing on fish acute toxicity across taxa. We analyzed the relevance of taxonomy and experimental setup, showing that taking them into account can lead to considerable improvements in the classification performance. We quantified the gain obtained throught the introduction of taxonomic and experimental information, compared to classification based on chemical information alone. We used our approach with standard machine learning models (K-nearest neighbors, random forests and deep neural networks), as well as the recently proposed Read-Across Structure Activity Relationship (RASAR) models, which were very successful in predicting chemical hazards to mammals based on chemical similarity. We were able to obtain accuracies of over 93% on datasets where, due to noise in the data, the maximum achievable accuracy was expected to be below 96%. The best performances were obtained by random forests and RASAR models. We analyzed metrics to compare our results with animal test reproducibility, and despite most of our models "outperform animal test reproducibility" as measured through recently proposed metrics, we showed that the comparison between machine learning performance and animal test reproducibility should be addressed with particular care. While we focused on fish mortality, our approach, provided that the right data is available, is valid for any combination of chemicals, effects and taxa.

We explore the use of policy gradient methods in reinforcement learning for quantum control via energy landscape shaping of XX-Heisenberg spin chains in a model agnostic fashion. Their performance is compared to finding controllers using gradient-based L-BFGS optimisation with restarts, with full access to an analytical model. Hamiltonian noise and coarse-graining of fidelity measurements are considered. Reinforcement learning is able to tackle challenging, noisy quantum control problems where L-BFGS optimization algorithms struggle to perform well. Robustness analysis under different levels of Hamiltonian noise indicates that controllers found by reinforcement learning appear to be less affected by noise than those found with L-BFGS., Comment: 7 pages, 7 figures

Advances in sensing and learning algorithms have led to increasingly mature solutions for human detection by robots, particularly in selected use-cases such as pedestrian detection for self-driving cars or close-range person detection in consumer settings. Despite this progress, the simple question "which sensor-algorithm combination is best suited for a person detection task at hand?" remains hard to answer. In this paper, we tackle this issue by conducting a systematic cross-modal analysis of sensor-algorithm combinations typically used in robotics. We compare the performance of state-of-the-art person detectors for 2D range data, 3D lidar, and RGB-D data as well as selected combinations thereof in a challenging industrial use-case. We further address the related problems of data scarcity in the industrial target domain, and that recent research on human detection in 3D point clouds has mostly focused on autonomous driving scenarios. To leverage these methodological advances for robotics applications, we utilize a simple, yet effective multi-sensor transfer learning strategy by extending a strong image-based RGB-D detector to provide cross-modal supervision for lidar detectors in the form of weak 3D bounding box labels. Our results show a large variance among the different approaches in terms of detection performance, generalization, frame rates and computational requirements. As our use-case contains difficulties representative for a wide range of service robot applications, we believe that these results point to relevant open challenges for further research and provide valuable support to practitioners for the design of their robot system., Comment: Accepted for publication at 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)