Search

SUMMARY Geophysical imaging using the inversion procedure is a powerful tool for the exploration of the Earth's subsurface. However, the interpretation of inverted images can sometimes be difficult, due to the inherent limitations of existing inversion algorithms, which produce smoothed sections. In order to improve and automate the processing and interpretation of inverted geophysical models, we propose an approach inspired from data mining. We selected an algorithm known as DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to perform clustering of inverted geophysical sections. The methodology relies on the automatic sorting and clustering of data. DBSCAN detects clusters in the inverted electrical resistivity values, with no prior knowledge of the number of clusters. This algorithm has the advantage of being defined by only two parameters: the neighbourhood of a point in the data space, and the minimum number of data points in this neighbourhood. We propose an objective procedure for the determination of these two parameters. The proof of concept described here is applied to simulated ERT (electrical resistivity tomography) sections, for the following three cases: two layers with a step, two layers with a rebound, and two layers with an anomaly embedded in the upper layer. To validate this approach, sensitivity studies were carried out on both of the above parameters, as well as to assess the influence of noise on the algorithm's performance. Finally, this methodology was tested on real field data. DBSCAN detects clusters in the inverted electrical resistivity models, and the former are then associated with various types of earth materials, thus allowing the structure of the prospected area to be determined. The proposed data-mining algorithm is shown to be effective, and to improve the interpretation of the inverted ERT sections. This new approach has considerable potential, as it can be applied to any geophysical data represented in the form of sections or maps.

Promising access to high-speed quantum networks relies on the creation of high-dimensional entangled memories that provide quantum communication with higher capacity of noisy quantum channels, thereby reducing the transmission time of information. Yet, the distribution of multidimensional entanglement between remote memory nodes is still faintly investigated. We propose an experimentally feasible protocol of deterministic generation of high-dimensional entanglement between distant multi-level atoms confined in high-finesse optical cavities and driven by laser pulses. Three-dimensional entanglement is generated deterministically between remote atoms by triggering a two-photon wavepacket, which mediates a superposition of states chosen by the laser pulse parameters. The efficient transfer of atomic states between remote nodes allows the construction of a three-dimensional quantum repeater, where the successful creation of entanglement can be verified by a simple method for reliable measurement of atomic ground Zeeman states., Comment: Accepted for publication at PRA. 10 pages, 5 figures

Estimating the precise location of a camera using visual localization enables interesting applications such as augmented reality or robot navigation. This is particularly useful in indoor environments where other localization technologies, such as GNSS, fail. Indoor spaces impose interesting challenges on visual localization algorithms: occlusions due to people, textureless surfaces, large viewpoint changes, low light, repetitive textures, etc. Existing indoor datasets are either comparably small or do only cover a subset of the mentioned challenges. In this paper, we introduce 5 new indoor datasets for visual localization in challenging real-world environments. They were captured in a large shopping mall and a large metro station in Seoul, South Korea, using a dedicated mapping platform consisting of 10 cameras and 2 laser scanners. In order to obtain accurate ground truth camera poses, we developed a robust LiDAR SLAM which provides initial poses that are then refined using a novel structure-from-motion based optimization. We present a benchmark of modern visual localization algorithms on these challenging datasets showing superior performance of structure-based methods using robust image features. The datasets are available at: https://naverlabs.com/datasets

We propose a new approach, based on the Hanbury Brown and Twiss intensity interferometry, to transform a Cherenkov telescope to its equivalent optical telescope. We show that, based on the use of photonics components borrowed from quantum-optical applications, we can recover spatial details of the observed source down to the diffraction limit of the Cherenkov telescope, set by its diameter at the mean wavelength of observation. For this, we propose to apply aperture synthesis techniques from pairwise and triple correlation of sub-pupil intensities, in order to reconstruct the image of a celestial source from its Fourier moduli and phase information, despite atmospheric turbulence. We examine the sensitivity of the method, i.e. limiting magnitude, and its implementation on existing or future high energy arrays of Cherenkov telescopes. We show that despite its poor optical quality compared to extremely large optical telescopes under construction, a Cherenkov telescope can provide diffraction limited imaging of celestial sources, in particular at the visible, down to violet wavelengths., Comment: 8 pages, 7 figures

We consider Taylor dispersion for tracer particles in micro-fluidic planar channels with strong confinement. In this context, the channel walls modify the local diffusivity tensor and also interactions between the tracer particles and the walls become important. We provide a simple and general formula for the effective diffusion constant along the channel as well as the first non-trivial finite time correction for arbitrary flows along the channel, arbitrary interaction potentials with the walls and arbitrary expressions for the diffusion tensor. The formula are in particular amenable to a straightforward numerical implementation, rendering them extremely useful for comparison with experiments. We present a number of applications, notably for systems which have parabolically varying diffusivity profiles, to systems with attractive interactions with the walls as well as electroosmotic flows between plates with differing surface charges within the Debye-H\"uckel approximation., Comment: 33 pages, 7 figures

One of the most fundamental open problems in physics is the unification of general relativity and quantum theory to a theory of quantum gravity. An aspect that might become relevant in such a theory is that the dynamical nature of causal structure present in general relativity displays quantum uncertainty. This may lead to a phenomenon known as indefinite or quantum causal structure, as captured by the process matrix formalism. Due to the generality of that framework, however, for many process matrices there is no clear physical interpretation. A popular approach towards a quantum theory of gravity is the Page-Wootters formalism, which associates to time a Hilbert space structure similar to spatial position. By explicitly introducing a quantum clock, it allows to describe time-evolution of systems via correlations between this clock and said systems encoded in history states. In this paper we combine the process matrix framework with a generalization of the Page-Wootters formalism in which one considers several agents, each with their own discrete quantum clock. We describe how to extract process matrices from scenarios involving such agents with quantum clocks, and analyze their properties. The description via a history state with multiple clocks imposes constraints on the implementation of process matrices and on the perspectives of the agents as described via causal reference frames. While it allows for scenarios where different definite causal orders are coherently controlled, we explain why certain noncausal processes might not be implementable within this setting., Comment: New title. Close to published version. Veronika Baumann and Marius Krumm share first authorship of this paper

In this work, we propose to extend a state-of-the-art multi-source localization system based on a convolutional recurrent neural network and Ambisonics signals. We significantly improve the performance of the baseline network by changing the layout between convolutional and pooling layers. We propose several configurations with more convolutional layers and smaller pooling sizes in-between, so that less information is lost across the layers, leading to a better feature extraction. In parallel, we test the system's ability to localize up to 3 sources, in which case the improved feature extraction provides the most significant boost in accuracy. We evaluate and compare these improved configurations on synthetic and real-world data. The obtained results show a quite substantial improvement of the multiple sound source localization performance over the baseline network., Comment: 5 pages, 2 figures. Accepted to EUSIPCO 2021

Unmanned Aerial Vehicles (UAVs) have the potential to be used for many applications in urban environments. However, allowing UAVs to fly above densely populated areas raises concerns regarding safety. One of the main safety issues is the possibility for a failure to cause the loss of navigation capabilities, which can result in the UAV falling/landing in hazardous areas such as busy roads, where it can cause fatal accidents. Current standards, such as the SORA published in 2019, do not consider applicable mitigation techniques to handle this kind of hazardous situations. Consequently, certifying UAV urban operations implies to demonstrate very high levels of integrity, which results in prohibitive development costs. To address this issue, this paper explores the concept of Emergency Landing (EL). A safety analysis is conducted on an urban UAV case study, and requirements are proposed to enable the integration of EL as an acceptable mitigation mean in the SORA. Based on these requirements, an EL implementation was developed, together with a runtime monitoring architecture to enhance confidence in the system. Preliminary qualitative results are presented and the monitor seem to be able to detect errors of the EL system effectively., Comment: 8 pages, 4 figure, 4 tables To appear in the proceedings of the 7th international workshop on Safety and Security of Intelligent Vehicles (SSIV 2021) at DSN 2021

Metaphorical expressions are difficult linguistic phenomena, challenging diverse Natural Language Processing tasks. Previous works showed that paraphrasing a metaphor as its literal counterpart can help machines better process metaphors on downstream tasks. In this paper, we interpret metaphors with BERT and WordNet hypernyms and synonyms in an unsupervised manner, showing that our method significantly outperforms the state-of-the-art baseline. We also demonstrate that our method can help a machine translation system improve its accuracy in translating English metaphors to 8 target languages.

We tackle the problem of identifying metaphors in text, treated as a sequence tagging task. The pre-trained word embeddings GloVe, ELMo and BERT have individually shown good performance on sequential metaphor identification. These embeddings are generated by different models, training targets and corpora, thus encoding different semantic and syntactic information. We show that leveraging GloVe, ELMo and feature-based BERT based on a multi-channel CNN and a Bidirectional LSTM model can significantly outperform any single word embedding method and the combination of the two embeddings. Incorporating linguistic features into our model can further improve model performance, yielding state-of-the-art performance on three public metaphor datasets. We also provide in-depth analysis on the effectiveness of leveraging multiple word embeddings, including analysing the spatial distribution of different embedding methods for metaphors and literals, and showing how well the embeddings complement each other in different genres and parts of speech.

The need to guarantee hard delay bounds to traffic flows with deterministic traffic profiles, e.g., token buckets, arises in a number of network settings. Of interest are solutions that offer such guarantees while minimizing network bandwidth. The paper explores a basic building block towards realizing such solutions, namely, a single hop configuration. The main results are in the form of optimal solutions for meeting local deadlines under schedulers of varying complexity and therefore cost. The results demonstrate how judiciously modifying flows' traffic profiles, i.e., reprofiling them, can help simple schedulers reduce the bandwidth they require, often performing nearly as well as more complex ones.

We report the formation of self-assembled monolayers of a molecular photoswitch (azobenzene-bithiophene derivative, AzBT) on cobalt via a thiol covalent bond. We study the electrical properties of the molecular junctions formed with the tip of a conductive atomic force microscope under ultra-high vacuum. The statistical analysis of the current-voltage curves shows two distinct states of the molecule conductance, suggesting the coexistence of both the trans and cis azobenzene isomers on the surface. The cis isomer population (trans isomer) increases (decreases) upon UV light irradiation. The situation is reversed under blue light irradiation. The experiments are confronted to first-principle calculations performed on the molecular junctions with the Non-Equilibrium Green's Function formalism combined with Density Functional Theory (NEGF/DFT). The theoretical results consider two different molecular orientations for each isomer. Whereas the orientation does not affect the conductance of the trans isomer, it significantly modulates the conductance of the cis isomer and the resulting conductance ON/OFF ratio of the molecular junction. This helps identifying the molecular orientation at the origin of the observed current differences between the trans and cis forms. The ON state is associated to the trans isomer irrespective of its orientation in the junction, while the OFF state is identified as a cis isomer with its azobenzene moiety folded upward with respect to the bithiophene core. The experimental and calculated ON/OFF conductance ratios have a similar order of magnitude. This conductance ratio seems reasonable to make these Co-AzBT molecular junctions a good test-bed to further explore the relationship between the spin-polarized charge transport, the molecule conformation and the molecule-Co spinterface., Comment: Full manuscript and supporting information

Artificial Intelligence systems cannot yet match human abilities to apply knowledge to situations that vary from what they have been programmed for, or trained for. In visual object recognition methods of inference exploiting top-down information (from a model) have been shown to be effective for recognising entities in difficult conditions. Here this type of inference, called `projection', is shown to be a key mechanism to solve the problem of applying knowledge to varied or challenging situations, across a range of AI domains, such as vision, robotics, or language. Finally the relevance of projection to tackling the commonsense knowledge problem is discussed., Comment: 29 pages, 3 figures. Some minor additions/clarifications in this revision, e.g. mathematical description

Object Detection (OD) is an important computer vision problem for industry, which can be used for quality control in the production lines, among other applications. Recently, Deep Learning (DL) methods have enabled practitioners to train OD models performing well on complex real world images. However, the adoption of these models in industry is still limited by the difficulty and the significant cost of collecting high quality training datasets. On the other hand, when applying OD to the context of production lines, CAD models of the objects to be detected are often available. In this paper, we introduce a fully automated method that uses a CAD model of an object and returns a fully trained OD model for detecting this object. To do this, we created a Blender script that generates realistic labeled datasets of images containing the object, which are then used for training the OD model. The method is validated experimentally on two practical examples, showing that this approach can generate OD models performing well on real images, while being trained only on synthetic images. The proposed method has potential to facilitate the adoption of object detection models in industry as it is easy to adapt for new objects and highly flexible. Hence, it can result in significant costs reduction, gains in productivity and improved products quality., Comment: 8 pages, 4 figures, 2 tables, To appear in the proceedings of the 23rd International Conference on Enterprise Information Systems (ICEIS 2021)

We study by scanning thermal microscopy the nanoscale thermal conductance of films (40 to 400 nm thick) of [1]benzothieno[3,2-b][1]benzothiophene (BTBT) and 2,7-dioctyl[1]benzothieno[3,2-b][1]benzothiophene (C8-BTBT-C8). We demonstrate that the out-of-plane thermal conductivity is significant along the interlayer direction, larger for BTBT (0.63 +/- 0.12 W m-1 K-1) compared to C8-BTBT-C8 (0.25 +/- 0.13 W m-1 K-1). These results are supported by molecular dynamics calculations (Approach to Equilibrium Molecular Dynamics method) performed on the corresponding molecular crystals. The calculations point to significant thermal conductivity (3D-like) values along the 3 crystalline directions, with anisotropy factors between the crystalline directions below 1.8 for BTBT and below 2.8 for C8-BTBT-C8, in deep contrast with the charge transport properties featuring a two-dimensional character for these materials. In agreement with the experiments, the calculations yield larger values in BTBT compared to C8-BTBT-C8 (0.6-1.3 W m-1 K-1 versus 0.3-0.7 W m-1 K-1, respectively). The weak thickness dependence of the nanoscale thermal resistance is in agreement with a simple analytical model., Comment: Nanoscale (2021)

We report on the installation and first results of one compact oceanographic radar in the region of Nice for a long-term observation of the coastal surface currents in the North-West Mediterranean Sea. We describe the specific processing and calibration techniques which were developed at the laboratory to produce high-quality radial surface current maps. In particular, we propose an original self-calibration technique of the antenna patterns, which is based on the sole analysis of the databasis and does not require any shipborne transponder or other external transmitters. The relevance of the self-calibration technique and the accuracy of inverted surface currents have been assessed with the launch of 40 drifters that remained under the radar coverage for about 10 days., Comment: 14 pages, 11 figures, 4 tables

Doppler radars are routinely used for the remote sensing of oceanic surface currents and atmospheric wind profiles. Even though they operate at different frequencies and address different media, they follow very similar processing for the extraction of measured velocities. In particular they both face the challenging issue of capturing geophysical phenomena which vary rapidly with respect to the typical integration time. Recently, the authors applied a non-spectral formalism based on autoregressive processes to model the backscattered time series obtained from High-Frequency oceanic radars. They showed that it allows to calculate Doppler spectra for very short integration times without losing in frequency resolution nor signal-to-noise ratio. We apply this technique to synthetic and experimental data within a common framework and show for the first time the strong potential of the method for the study of transient atmospheric phenomena., Comment: 4 pages, 3 figures submitted for publication to the IEEE International Geoscience and Remote Sensing Symposium (IGARSS), Brussels, Belgium, July 11-16, 2021

Guided by the intuition of coherent superposition of causal relations, recent works presented quantum processes without classical common-cause and direct-cause explanation, that is, processes which cannot be written as probabilistic mixtures of quantum common-cause and quantum direct-cause relations (CCDC). In this work, we analyze the minimum requirements for a quantum process to fail to admit a CCDC explanation and present "simple" processes, which we prove to be the most robust ones against general noise. These simple processes can be realized by preparing a maximally entangled state and applying the identity quantum channel, thus not requiring an explicit coherent mixture of common-cause and direct-cause, exploiting the possibility of a process to have both relations simultaneously. We then prove that, although all bipartite direct-cause processes are bipartite separable operators, there exist bipartite separable processes which are not direct-cause. This shows that the problem of deciding weather a process is direct-cause process is not equivalent to entanglement certification and points out the limitations of entanglement methods to detect non-classical CCDC processes. We also present a semi-definite programming hierarchy that can detect and quantify the non-classical CCDC robustnesses of every non-classical CCDC process. Among other results, our numerical methods allow us to show that the simple processes presented here are likely to be also the maximally robust against white noise. Finally, we explore the equivalence between bipartite direct-cause processes and bipartite processes without quantum memory, to present a separable process which cannot be realized as a process without quantum memory., Comment: 24+8 pages, 11 figures. Code available at https://github.com/marcellongvb/non_ccdc_processes

The cloud's flexibility and promise of seamless auto-scaling notwithstanding, its ability to meet service level objectives (SLOs) typically calls for some form of control in resource usage. This seemingly traditional problem gives rise to new challenges in a cloud setting, and in particular a subtle yet significant trade-off involving load-distribution decisions (the distribution of workload across available cloud resources to optimize performance), and rate limiting (the capping of individual workloads to prevent global over-commitment). This paper investigates that trade-off through the design and implementation of a real-time messaging system motivated by Internet-of-Things (IoT) applications, and demonstrates a solution capable of realizing an effective compromise. The paper's contributions are in both explicating the source of this trade-off, and in demonstrating a possible solution.

Recently, a common starting point for solving complex unsupervised image classification tasks is to use generic features, extracted with deep Convolutional Neural Networks (CNN) pretrained on a large and versatile dataset (ImageNet). However, in most research, the CNN architecture for feature extraction is chosen arbitrarily, without justification. This paper aims at providing insight on the use of pretrained CNN features for image clustering (IC). First, extensive experiments are conducted and show that, for a given dataset, the choice of the CNN architecture for feature extraction has a huge impact on the final clustering. These experiments also demonstrate that proper extractor selection for a given IC task is difficult. To solve this issue, we propose to rephrase the IC problem as a multi-view clustering (MVC) problem that considers features extracted from different architectures as different "views" of the same data. This approach is based on the assumption that information contained in the different CNN may be complementary, even when pretrained on the same data. We then propose a multi-input neural network architecture that is trained end-to-end to solve the MVC problem effectively. This approach is tested on nine natural image datasets, and produces state-of-the-art results for IC., Comment: 21 pages, 16 figures, 10 tables, preprint of our paper published in Neurocomputing

Speaker counting is the task of estimating the number of people that are simultaneously speaking in an audio recording. For several audio processing tasks such as speaker diarization, separation, localization and tracking, knowing the number of speakers at each timestep is a prerequisite, or at least it can be a strong advantage, in addition to enabling a low latency processing. In a previous work, we addressed the speaker counting problem with a multichannel convolutional recurrent neural network which produces an estimation at a short-term frame resolution. In this work, we show that, for a given frame, there is an optimal position in the input sequence for best prediction accuracy. We empirically demonstrate the link between that optimal position, the length of the input sequence and the size of the convolutional filters., Comment: Presented at Forum Acusticum 2020

We study numerically the slow (subradiant) decay of the fluorescence of motionless atoms after a weak pulsed excitation. We show that, in the linear-optics regime and for an excitation detuned by several natural linewidths, the slow decay rate can be dominated by close pairs of atoms (dimers) forming superradiant and subradiant states. However, for a large-enough resonant optical depth and at later time, the dynamics is dominated by collective many-body effects. In this regime, we study the polarization and the spectrum of the emitted light, as well as the spatial distribution of excitation inside the sample, as a function of time during the decay dynamics. The behavior of these observables is consistent with what would be expected for radiation trapping of nearly resonant light. This finding sheds light on subradiance in dilute samples by providing an interpretation based on the light behavior of the system (multiple scattering) which is complementary to the more commonly used picture of the collective atomic Dicke state.

We investigate numerically the role of near-field dipole-dipole interactions on the late emission dynamics of large disordered cold atomic samples driven by a weak field. Previous experimental and numerical studies of subradiance in macroscopic samples have focused on low-density samples of pure two-level atoms, without internal structure, which corresponds to a scalar representation of the light. The cooperative nature of the late emission of light is then governed by the resonant optical depth. Here, by considering the vectorial nature of the light, we show the detrimental role of the near-field terms on cooperativity in higher-density samples. The observed reduction in the subradiant lifetimes is interpreted as a signature of the inhomogeneous broadening due to the near-field contributions, in analogy with the Van der Waals dephasing phenomenon for superradiance.

Visual storytelling is a creative and challenging task, aiming to automatically generate a story-like description for a sequence of images. The descriptions generated by previous visual storytelling approaches lack coherence because they use word-level sequence generation methods and do not adequately consider sentence-level dependencies. To tackle this problem, we propose a novel hierarchical visual storytelling framework which separately models sentence-level and word-level semantics. We use the transformer-based BERT to obtain embeddings for sentences and words. We then employ a hierarchical LSTM network: the bottom LSTM receives as input the sentence vector representation from BERT, to learn the dependencies between the sentences corresponding to images, and the top LSTM is responsible for generating the corresponding word vector representations, taking input from the bottom LSTM. Experimental results demonstrate that our model outperforms most closely related baselines under automatic evaluation metrics BLEU and CIDEr, and also show the effectiveness of our method with human evaluation.