Search

Airborne topographic LiDAR is an active remote sensing technology that emits near-infrared light to map objects on the Earth's surface. Derived products of LiDAR are suitable to service a wide range of applications because of their rich three-dimensional spatial information and their capacity to obtain multiple returns. However, processing point cloud data still requires a significant effort in manual editing. Certain human-made objects are difficult to detect because of their variety of shapes, irregularly-distributed point clouds, and low number of class samples. In this work, we propose an efficient end-to-end deep learning framework to automatize the detection and segmentation of objects defined by an arbitrary number of LiDAR points surrounded by clutter. Our method is based on a light version of PointNet that achieves good performance on both object recognition and segmentation tasks. The results are tested against manually delineated power transmission towers and show promising accuracy., Comment: proceedings of the 24th International Conference of the Catalan Association for Artificial Intelligence (CCIA 2022)

The detection of GW170817 and the accompanying electromagnetic counterpart, AT2017gfo, have provided an important set of observational constraints for theoretical models of neutron star mergers, nucleosynthesis, and radiative transfer for kilonovae. We apply the 3D Monte Carlo radiative transfer code ARTIS to produce synthetic light curves of the dynamical ejecta from a neutron star merger, which has been modelled with 3D smooth-particle hydrodynamics (SPH) and included neutrino interactions. Nucleosynthesis calculations provide the energy released from radioactive decays of r-process nuclei, and radiation transport is performed using grey opacities given as functions of the electron fraction. We present line-of-sight dependent bolometric light curves, and find the emission along polar lines of sight to be up to a factor of ~2 brighter than along equatorial lines of sight. Instead of a distinct emission peak, our bolometric light curve exhibits a monotonic decline, characterised by a shoulder at the time when the bulk ejecta becomes optically thin. We show approximate band light curves based on radiation temperatures and compare these to the observations of AT2017gfo. We find that the rapidly declining temperatures lead to a blue to red colour evolution similar to that shown by AT2017gfo. We also investigate the impact of an additional, spherically symmetric secular ejecta component, and we find that the early light curve remains nearly unaffected, while after about 1 day the emission is strongly enhanced and dominated by the secular ejecta, leading to the shift of the shoulder from 1-2 to 6-10 days., Comment: Accepted by MNRAS

We present a generalised version of the semi-analytic Just-Jahreiss (JJ) model of the Galactic disk that incorporates our findings for the solar neighbourhood and is applicable to a wide range of galactocentric distances, 4 kpc $\lesssim R \lesssim $ 14 kpc. The JJ model is a flexible tool for stellar population synthesis with a fine age resolution of 25 Myr. It includes six Milky-Way (MW) components: the four flattened (exponential thin and thick disk, atomic and molecular gas) and two spheroidal (spherical stellar halo and a cored isothermal dark matter sphere). The overall thin-disk thickness is assumed to be constant at all radii, though flaring can also be tested. The adopted radial variation in the thin-disk star-formation rate (SFR) reflects the inside-out disk growth scenario. We allow a smooth power-law SFR continuum to be modified by an arbitrary number of Gaussian peaks. We present a public code of the JJ model complemented by the three sets of isochrones (PARSEC, MIST, and BaSTI). Using metallicity distributions of the red clump giants from APOGEE, we constrain the radial variation of the JJ-model age-metallicity relation (AMR) and propose a new analytic form for the AMR function. The generalised JJ model is a publicly available tool for studying different stellar populations across the MW disk. With its fine age resolution and flexibility, it can be particularly useful for reconstructing the thin-disk SFR, as a variety of different SFR shapes can be constructed within its framework., Comment: 21 pages, 12 figures, 3 tables. Accepted for publication by A&A

Theoretically predicted yields of elements created by the rapid neutron capture (r-) process carry potentially large uncertainties associated with incomplete knowledge of nuclear properties and approximative hydrodynamical modelling of the matter ejection processes. We present an in-depth study of the nuclear uncertainties by varying theoretical nuclear input models that describe the experimentally unknown neutron-rich nuclei. This includes two frameworks for calculating the radiative neutron capture rates and 14 different models for nuclear masses, $\beta$-decay rates and fission properties. Our r-process nuclear network calculations are based on detailed hydrodynamical simulations of dynamically ejected material from NS-NS or NS-BH binary mergers plus the secular ejecta from BH-torus systems. The impact of nuclear uncertainties on the r-process abundance distribution and the early radioactive heating rate is found to be modest (within a factor of $\sim20$ for individual $A>90$ abundances and a factor of 2 for the heating rate). However, the impact on the late-time heating rate is more significant and depends strongly on the contribution from fission. We witness significantly larger sensitivity to the nuclear physics input if only a single trajectory is used compared to considering ensembles of $\sim$200-300 trajectories, and the quantitative effects of the nuclear uncertainties strongly depend on the adopted conditions for the individual trajectory. We use the predicted Th/U ratio to estimate the cosmochronometric age of six metal-poor stars and find the impact of the nuclear uncertainties to be up to 2 Gyr., Comment: 28 pages, 24 figures, accepted to MNRAS

Today, companies and data centers are moving towards cloud and serverless storage systems instead of traditional file systems. As a result of such a transition, allocating sufficient resources to users and parties to satisfy their service level demands has become crucial in cloud storage. In cloud storage, the schedulability of system components and requests is of great importance to achieving QoS goals. However, the bufferbloat phenomenon in storage backends impacts the schedulability of the system. In a storage server, bufferbloat happens when the server submits all requests immediately to the storage backend due to a large buffer in the backend. In recent decades, many studies have focused on the bufferbloat as a latency problem. Nevertheless, none of these works investigate the impact of bufferbloat on the schedulability of the system. In this paper, we demonstrate that the bufferbloat impacts scheduling and performance isolation and identify utilizing admission control in the storage backend as an easy-to-adopt solution to mitigate bufferbloat. Moreover, we show that traditional static admission controls are inadequate in the face of dynamic workloads in cloud environments. Finally, we propose SlowFast CoDel, an adaptive admission control, as a starting point for developing adaptive admission control mechanisms to mitigate bufferbloat in cloud storage.

Collapsar disks have been proposed to be rich factories of heavy elements, but the major question of whether their outflows are neutron-rich, and could therefore represent significant sites of the rapid neutron-capture (r-) process, or dominated by iron-group elements remains unresolved. We present the first global models of collapsars that start from a stellar progenitor and self-consistently describe the evolution of the disk, its composition, and its outflows in response to the imploding stellar mantle, using energy-dependent M1 neutrino transport and an alpha-viscosity to approximate turbulent angular-momentum transport. We find that a neutron-rich, neutrino-dominated accretion flow (NDAF) is established only marginally--either for short times or relatively low viscosities--because the disk tends to disintegrate into an advective disk (ADAF) already at relatively high mass-accretion rates, launching powerful outflows but preventing it from developing a hot, dense, and therefore neutron-rich core. Viscous outflows disrupt the star within ~100s with explosion energies close to that of hypernovae. If viscosity is neglected, a stable NDAF with disk mass of about 1Msun is formed but is unable to release neutron-rich ejecta, while it produces a relatively mild explosion powered by a neutrino-driven wind blown off its surface. With ejecta electron fractions close to 0.5, all models presumably produce large amounts of Ni56. Our results suggest that collapsar models based on the alpha-viscosity are inefficient r-process sites and that genuinely magnetohydrodynamic effects may be required to generate neutron-rich outflows. A relatively weak effective viscosity generated by magnetohydrodynamic turbulence would improve the prospects for obtaining neutron-rich ejecta., Comment: 7 pages, 4 figures, 1 table, slightly revised discussion, main results unchanged compared to v1, accepted to ApJL

Photo- and charge-carrier induced ion migration is a major challenge when utilizing metal halide perovskite semiconductors for optoelectronic applications. For mixed iodide/bromide perovskites, the compositional instability due to light- or electrical bias induced phase- segregation restricts the exploitation of the entire bandgap range. Previous experimental and theoretical work suggests that excited states or charge-carriers trigger the process but the exact mechanism is still under debate. To identify the mechanism and cause of light-induced phase-segregation phenomena we investigate the full compositional range of methylammonium lead bromide/iodide samples, MAPb(Br$_x$I$_{1-x}$)$_3$ with $x = 0\ldots 1$, by simultaneous in-situ X-ray diffraction and photoluminescence spectroscopy during illumination. The quantitative comparison of composition-dependent in-situ XRD and PL shows that at excitation densities of 1 sun, only the initial stage of photo-segregation can be rationalized with the previously established thermodynamic models. However, we observe a progression of the phase-segregation that can only be rationalized by considering long-lived accumulative photo-induced material alterations. We suggest that (additional) photo-induced defects, possibly halide vacancies and interstitials, need to be considered to fully rationalize light-induced phase-segregation and anticipate our findings to provide crucial insight for the development of more sophisticated models.

Backdoor attacks insert malicious data into a training set so that, during inference time, it misclassifies inputs that have been patched with a backdoor trigger as the malware specified label. For backdoor attacks to bypass human inspection, it is essential that the injected data appear to be correctly labeled. The attacks with such property are often referred to as "clean-label attacks." Existing clean-label backdoor attacks require knowledge of the entire training set to be effective. Obtaining such knowledge is difficult or impossible because training data are often gathered from multiple sources (e.g., face images from different users). It remains a question whether backdoor attacks still present a real threat. This paper provides an affirmative answer to this question by designing an algorithm to mount clean-label backdoor attacks based only on the knowledge of representative examples from the target class. With poisoning equal to or less than 0.5% of the target-class data and 0.05% of the training set, we can train a model to classify test examples from arbitrary classes into the target class when the examples are patched with a backdoor trigger. Our attack works well across datasets and models, even when the trigger presents in the physical world. We explore the space of defenses and find that, surprisingly, our attack can evade the latest state-of-the-art defenses in their vanilla form, or after a simple twist, we can adapt to the downstream defenses. We study the cause of the intriguing effectiveness and find that because the trigger synthesized by our attack contains features as persistent as the original semantic features of the target class, any attempt to remove such triggers would inevitably hurt the model accuracy first., Comment: 13 pages of the main text

The outflows from neutrino-cooled black-hole (BH) accretion disks formed in neutron-star mergers or cores of collapsing stars are expected to be neutron-rich enough to explain a large fraction of elements created by the rapid neutron-capture (r-) process, but their precise chemical composition remains elusive. Here, we investigate the role of fast neutrino flavor conversion, motivated by the findings of our post-processing analysis that shows evidence of electron-neutrino lepton-number (ELN) crossings deep inside the disk, hence suggesting possibly non-trivial effects due to neutrino flavor mixing. We implement a parametric, dynamically self-consistent treatment of fast conversion in time-dependent simulations and examine the impact on the disk and its outflows. By activating the, otherwise inefficient, emission of heavy-lepton neutrinos, fast conversions enhance the disk cooling rates and reduce the absorption rates of electron-type neutrinos, causing a reduction of the electron fraction in the disk by 0.03-0.06 and in the ejected material by 0.01-0.03. The r-process yields are enhanced by typically no more than a factor of two, rendering the overall impact of fast conversions modest. The kilonova is prolonged as a net result of increased lanthanide opacities and enhanced radioactive heating rates. We observe only mild sensitivity to the disk mass, the condition for the onset of flavor conversion, and to the considered cases of flavor mixing. Remarkably, parametric models of flavor mixing that conserve the lepton numbers per family result in an overall smaller impact than models invoking three-flavor equipartition, often assumed in previous works., Comment: 25 pages, 10 figures, 1 table; accepted to PRD

Brain-computer interfaces (BCIs) decode recorded neural signals from the brain and/or stimulate the brain with encoded neural signals. BCIs span both hardware and software and have a wide range of applications in restorative medicine, from restoring movement through prostheses and robotic limbs to restoring sensation and communication through spellers. BCIs also have applications in diagnostic medicine, e.g., providing clinicians with data for detecting seizures, sleep patterns, or emotions. Despite their promise, BCIs have not yet been adopted for long-term, day-to-day use because of challenges related to reliability and robustness, which are needed for safe operation in all scenarios. Ensuring safe operation currently requires hours of manual data collection and recalibration, involving both patients and clinicians. However, data collection is not targeted at eliminating specific faults in a BCI. This paper presents a new methodology for characterizing, detecting, and localizing faults in BCIs. Specifically, it proposes partial test oracles as a method for detecting faults and slice functions as a method for localizing faults to characteristic patterns in the input data or relevant tasks performed by the user. Through targeted data acquisition and retraining, the proposed methodology improves the correctness of BCIs. We evaluated the proposed methodology on five BCI applications. The results show that the proposed methodology (1) precisely localizes faults and (2) can significantly reduce the frequency of faults through retraining based on targeted, fault-based data acquisition. These results suggest that the proposed methodology is a promising step towards repairing faulty BCIs., Comment: Accepted at International Conference on Software Engineering (ICSE-2022)

This paper presents a framework to perform bifurcation analysis in laboratory experiments or simulations. We employ control-based continuation to study the dynamics of a macroscopic variable of a microscopically defined model, exploring the potential viability of the underlying feedback control techniques in an experiment. In contrast to previous experimental studies that used iterative root-finding methods on the feedback control targets, we propose a feedback control law that is inherently non-invasive. That is, the control discovers the location of equilibria and stabilizes them simultaneously. We call the proposed control zero-in-equilibrium feedback control and we prove that it is able to stabilize branches of equilibria, except at singularities of codimension n+1, where n is the number of state space dimensions the feedback can depend on. We apply the method to a simulated evacuation scenario were pedestrians have to reach an exit after maneuvering left or right around an obstacle. The scenario shows a hysteresis phenomenon with bistability and tipping between two possible steady pedestrian flows in microscopic simulations. We demonstrate for the evacuation scenario that the proposed control law is able to uniformly discover and stabilize steady flows along the entire branch, including points where other non-invasive approaches to feedback control become singular., Comment: submitted (34 pages 12 figures)

Recent studies show that the state-of-the-art deep neural networks are vulnerable to model inversion attacks, in which access to a model is abused to reconstruct private training data of any given target class. Existing attacks rely on having access to either the complete target model (whitebox) or the model's soft-labels (blackbox). However, no prior work has been done in the harder but more practical scenario, in which the attacker only has access to the model's predicted label, without a confidence measure. In this paper, we introduce an algorithm, Boundary-Repelling Model Inversion (BREP-MI), to invert private training data using only the target model's predicted labels. The key idea of our algorithm is to evaluate the model's predicted labels over a sphere and then estimate the direction to reach the target class's centroid. Using the example of face recognition, we show that the images reconstructed by BREP-MI successfully reproduce the semantics of the private training data for various datasets and target model architectures. We compare BREP-MI with the state-of-the-art whitebox and blackbox model inversion attacks and the results show that despite assuming less knowledge about the target model, BREP-MI outperforms the blackbox attack and achieves comparable results to the whitebox attack., Comment: Accepted at CVPR2022

Proper statistical modeling incorporates domain theory about how concepts relate and details of how data were measured. However, data analysts currently lack tool support for recording and reasoning about domain assumptions, data collection, and modeling choices in an integrated manner, leading to mistakes that can compromise scientific validity. For instance, generalized linear mixed-effects models (GLMMs) help answer complex research questions, but omitting random effects impairs the generalizability of results. To address this need, we present Tisane, a mixed-initiative system for authoring generalized linear models with and without mixed-effects. Tisane introduces a study design specification language for expressing and asking questions about relationships between variables. Tisane contributes an interactive compilation process that represents relationships in a graph, infers candidate statistical models, and asks follow-up questions to disambiguate user queries to construct a valid model. In case studies with three researchers, we find that Tisane helps them focus on their goals and assumptions while avoiding past mistakes.

Learning effective visual representations without human supervision is a long-standing problem in computer vision. Recent advances in self-supervised learning algorithms have utilized contrastive learning, with methods such as SimCLR, which applies a composition of augmentations to an image, and minimizes a contrastive loss between the two augmented images. In this paper, we present CLAWS, an annotation-efficient learning framework, addressing the problem of manually labeling large-scale agricultural datasets along with potential applications such as anomaly detection and plant growth analytics. CLAWS uses a network backbone inspired by SimCLR and weak supervision to investigate the effect of contrastive learning within class clusters. In addition, we inject a hard attention mask to the cropped input image before maximizing agreement between the image pairs using a contrastive loss function. This mask forces the network to focus on pertinent object features and ignore background features. We compare results between a supervised SimCLR and CLAWS using an agricultural dataset with 227,060 samples consisting of 11 different crop classes. Our experiments and extensive evaluations show that CLAWS achieves a competitive NMI score of 0.7325. Furthermore, CLAWS engenders the creation of low dimensional representations of very large datasets with minimal parameter tuning and forming well-defined clusters, which lends themselves to using efficient, transparent, and highly interpretable clustering methods such as Gaussian Mixture Models.

In this work, we propose ModelPred, a framework that helps to understand the impact of changes in training data on a trained model. This is critical for building trust in various stages of a machine learning pipeline: from cleaning poor-quality samples and tracking important ones to be collected during data preparation, to calibrating uncertainty of model prediction, to interpreting why certain behaviors of a model emerge during deployment. Specifically, ModelPred learns a parameterized function that takes a dataset $S$ as the input and predicts the model obtained by training on $S$. Our work differs from the recent work of Datamodels [1] as we aim for predicting the trained model parameters directly instead of the trained model behaviors. We demonstrate that a neural network-based set function class is capable of learning the complex relationships between the training data and model parameters. We introduce novel global and local regularization techniques to prevent overfitting and we rigorously characterize the expressive power of neural networks (NN) in approximating the end-to-end training process. Through extensive empirical investigations, we show that ModelPred enables a variety of applications that boost the interpretability and accountability of machine learning (ML), such as data valuation, data selection, memorization quantification, and model calibration.

Cognitive neuroscience methods can identify the fMRI-measured neural representation of familiar individual concepts, such as apple, and decompose them into meaningful neural and semantic components. This approach was applied here to determine the neural representations and underlying dimensions of representation of far more abstract physics concepts related to matter and energy, such as fermion and dark matter, in the brains of 10 Carnegie Mellon physics faculty members who thought about the main properties of each of the concepts. One novel dimension coded the measurability vs. immeasurability of a concept. Another novel dimension of representation evoked particularly by post-classical concepts was associated with four types of cognitive processes, each linked to particular brain regions: (1) Reasoning about intangibles, taking into account their separation from direct experience and observability; (2) Assessing consilience with other, firmer knowledge; (3) Causal reasoning about relations that are not apparent or observable; and (4) Knowledge management of a large knowledge organization consisting of a multi-level structure of other concepts. Two other underlying dimensions, previously found in physics students, periodicity, and mathematical formulation, were also present in this faculty sample. The data were analyzed using factor analysis of stably responding voxels, a Gaussian-na\"ive Bayes machine-learning classification of the activation patterns associated with each concept, and a regression model that predicted activation patterns associated with each concept based on independent ratings of the dimensions of the concepts. The findings indicate that the human brain systematically organizes novel scientific concepts in terms of new dimensions of neural representation., Comment: 12 pages, 6 figures, open access publication

The majority of existing results for the kilonova (or macronova) emission from material ejected during a neutron-star (NS) merger is based on (quasi-)one-zone models or manually constructed toy-model ejecta configurations. In this study we present a kilonova analysis of the material ejected during the first ~10ms of a NS merger, called dynamical ejecta, using directly the outflow trajectories from general relativistic smoothed-particle hydrodynamics simulations including a sophisticated neutrino treatment and the corresponding nucleosynthesis results, which have been presented in Part I of this study. We employ a multi-dimensional two-moment radiation transport scheme with approximate M1 closure to evolve the photon field and use a heuristic prescription for the opacities found by calibration with atomic-physics based reference results. We find that the photosphere is generically ellipsoidal but augmented with small-scale structure and produces emission that is about 1.5-3 times stronger towards the pole than the equator. The kilonova typically peaks after 0.7-1.5days in the near-infrared frequency regime with luminosities between 3-7x10^40erg/s and at photospheric temperatures of 2.2-2.8x10^3K. A softer equation of state or higher binary-mass asymmetry leads to a longer and brighter signal. Significant variations of the light curve are also obtained for models with artificially modified electron fractions, emphasizing the importance of a reliable neutrino-transport modeling. None of the models investigated here, which only consider dynamical ejecta, produces a transient as bright as AT2017gfo. The near-infrared peak of our models is incompatible with the early blue component of AT2017gfo., Comment: 23 pages, 15 figures, 1 table, accepted to MNRAS; updated for correct version of Fig. A1, left panel, and corrected Eqs. 7 and 9

Unsupervised disentangled representation learning is a long-standing problem in computer vision. This work proposes a novel framework for performing image clustering from deep embeddings by combining instance-level contrastive learning with a deep embedding based cluster center predictor. Our approach jointly learns representations and predicts cluster centers in an end-to-end manner. This is accomplished via a three-pronged approach that combines a clustering loss, an instance-wise contrastive loss, and an anchor loss. Our fundamental intuition is that using an ensemble loss that incorporates instance-level features and a clustering procedure focusing on semantic similarity reinforces learning better representations in the latent space. We observe that our method performs exceptionally well on popular vision datasets when evaluated using standard clustering metrics such as Normalized Mutual Information (NMI), in addition to producing geometrically well-separated cluster embeddings as defined by the Euclidean distance. Our framework performs on par with widely accepted clustering methods and outperforms the state-of-the-art contrastive learning method on the CIFAR-10 dataset with an NMI score of 0.772, a 7-8% improvement on the strong baseline.

Aims: We study the impact of Post-Newtonian correction terms on the energetic interaction between a gravitational wave (GW)-emitting supermassive black hole (SMBH) binary system and incoming stars via three-body scattering experiments. Methods: We use the AR-chain code to simulate with high accuracy the interactions between stars and an SMBH binary at separations of$\sim 1000\,R_{sch}$. For all of the interactions, we investigate in detail the energy balance of the three-body systems, using both Newtonian and Post-Newtonian expressions for the SMBH binary orbital energy, taking into account the GW emission by the binary. Results: We find that at these separations, purely Newtonian treatment of the binary orbital energy is insufficient to properly account for the SMBH binary orbital evolution. Instead, along with GW emission, even terms in the PN-corrections must be included in order to describe the energy change of the binary during the stellar interaction.es, 2, Comment: 8 pages, 2 figures, 1 table. Submitted to A&A

We present a coherent study of the impact of neutrino interactions on the r-process element nucleosynthesis and the heating rate produced by the radioactive elements synthesised in the dynamical ejecta of neutron star-neutron star (NS-NS) mergers. We have studied the material ejected from four NS-NS merger systems based on hydrodynamical simulations which handle neutrino effects in an elaborate way by including neutrino equilibration with matter in optically thick regions and re-absorption in optically thin regions. We find that the neutron richness of the dynamical ejecta is significantly affected by the neutrinos emitted by the post-merger remnant, in particular when compared to a case neglecting all neutrino interactions. Our nucleosynthesis results show that a solar-like distribution of r- process elements with mass numbers $A \gtrsim 90$ is produced, including a significant enrichment in Sr and a reduced production of actinides compared to simulations without inclusion of the nucleonic weak processes. The composition of the dynamically ejected matter as well as the corresponding rate of radioactive decay heating are found to be rather independent of the system mass asymmetry and the adopted equation of state. This approximate degeneracy in abundance pattern and heating rates can be favourable for extracting the ejecta properties from kilonova observations, at least if the dynamical component dominates the overall ejecta. Part II of this work will study the light curve produced by the dynamical ejecta of our four NS merger models., Comment: 18 pages, 21 figures, 2 Tables, accepted by MNRAS 2021-11-19