Recent Submissions

  • Interaction of Dust Aerosols with Land/Sea Breezes over the Eastern Coast of the Red Sea from LIDAR Data and High-resolution WRF-Chem Simulations

    Parajuli, Sagar P.; Stenchikov, Georgiy L.; Ukhov, Alexander; Shevchenko, Illia; Dubovik, Oleg; Lopatin, Anton (Submitted to Atmospheric Chemistry and Physics Discussions, Copernicus GmbH, 2020-07-08) [Preprint]
    With advances in modeling approaches and the application of satellite and ground-based data in dust-related research, our understanding of the dust cycle has significantly improved in recent decades. However, two aspects of the dust cycle, namely the vertical profiles and diurnal cycles, are not yet adequately understood, mainly due to the sparsity of direct observations. Measurements of backscattering caused by atmospheric aerosols have been ongoing since 2014 at the King Abdullah University of Science and Technology (KAUST) campus using a micro-pulse LIDAR with a high temporal resolution. KAUST is located on the east coast of the Red Sea (22.3° N, 39.1° E), and currently hosts the only operating LIDAR system in the Arabian Peninsula. We use the data from this LIDAR together with other collocated observations and high-resolution WRF-Chem model simulations to study the following aspects of aerosols, with a focus on dust over the Red Sea Arabian coastal plains. Firstly, we investigate the vertical profiles of aerosol extinction and concentration in terms of their seasonal and diurnal variability. Secondly, we evaluate how well the WRF-Chem model performs in representing the vertical distribution of aerosols over the study site. Thirdly, we explore the interactions between dust aerosols and land/sea breezes, which are the most influential components of the local diurnal circulation in the region. We found a substantial variation in the vertical profile of aerosols in different seasons. We also discovered a marked difference in the daytime and nighttime vertical distribution of aerosols at the study site, as revealed by the LIDAR data. The LIDAR data also identified a prominent dust layer at ∼5–7 km during the nighttime, which represented the long-range transported dust brought to the site by the easterly flow from remote inland deserts. The vertical profiles of aerosol extinction in different seasons were largely consistent between the LIDAR, MERRA-2 reanalysis, and CALIOP data, as well as in the WRF-Chem simulations. The sea breeze circulation was much deeper (∼2 km) than the land breeze circulation (∼1 km), but both breeze systems prominently affected the distribution of dust aerosols over the study site. We observed that sea breezes push the dust aerosols upwards along the western slope of the Sarawat Mountains, which eventually collide with the dust-laden northeasterly trade winds coming from nearby inland deserts, causing elevated dust maxima at a height of ∼1.5 km above sea level over the mountains. Moreover, the sea and land breezes intensified dust emissions from the coastal region during the daytime and nighttime, respectively. The WRF-Chem model successfully captured the onset, demise, and height of a large-scale dust event that occurred in 2015, compared to LIDAR data. Our study, although focused on a particular region, has broader environmental implications as it highlights how aerosols and dust emissions from the coastal plains can affect the Red Sea climate and marine habitats.
  • Semiparametric estimation of cross-covariance functions for multivariate random fields

    Qadir, Ghulam A.; Sun, Ying (Biometrics, Wiley, 2020-07-06) [Article]
    The prevalence of spatially referenced multivariate data has impelled researchers to develop procedures for joint modeling of multiple spatial processes. This ordinarily involves modeling marginal and cross-process dependence for any arbitrary pair of locations using a multivariate spatial covariance function. However, building a flexible multivariate spatial covariance function that is nonnegative definite is challenging. Here, we propose a semiparametric approach for multivariate spatial covariance function estimation with approximate Matérn marginals and highly flexible cross-covariance functions via their spectral representations. The flexibility in our cross-covariance function arises due to B-spline based specification of the underlying coherence functions, which in turn allows us to capture non-trivial cross-spectral features. We then develop a likelihood-based estimation procedure and perform multiple simulation studies to demonstrate the performance of our method, especially on the coherence function estimation. Finally, we analyze particulate matter concentrations (PM2.5) and wind speed data over the West-North-Central climatic region of the United States, where we illustrate that our proposed method outperforms the commonly used full bivariate Matérn model and the linear model of coregionalization for spatial prediction.
  • Simultaneous Bayesian Estimation of Non-Planar Fault Geometry and Spatially-Variable Slip

    Dutta, Rishabh; Jonsson, Sigurjon; Vasyura-Bathke, Hannes (Wiley, 2020-07-02) [Preprint]
    Large earthquakes are usually modeled with simple planar fault surfaces or a combination of several planar fault segments. However, in general, earthquakes occur on faults that are non-planar and exhibit significant geometrical variations in both the along-strike and down-dip directions at all spatial scales. Mapping of surface fault ruptures and high-resolution geodetic observations are increasingly revealing complex fault geometries near the surface and accurate locations of aftershocks often indicate geometrical complexities at depth. With better geodetic data and observations of fault ruptures, more details of complex fault geometries can be estimated resulting in more realistic fault models of large earthquakes. To address this topic, we here parametrize non-planar fault geometries with a set of polynomial parameters that allow for both along-strike and down-dip variations in the fault geometry. Our methodology uses Bayesian inference to estimate the non-planar fault parameters from geodetic data, yielding an ensemble of plausible models that characterize the uncertainties of the non-planar fault geometry and the fault slip. The method is demonstrated using synthetic tests considering checkerboard fault-slip patterns on non-planar fault surfaces with spatially-variable dip and strike angles both in the down-dip and in the along-strike directions. The results show that fault-slip estimations can be biased when a simple planar fault geometry is assumed in presence of significant non-planar geometrical variations. Our method can help to model earthquake fault sources in a more realistic way and may be extended to include multiple non-planar fault segments or other geometrical fault complexities.
  • A Robust, Safe and Scalable Magnetic Nanoparticle Workflow for RNA Extraction of Pathogens from Clinical and Environmental Samples

    Ramos Mandujano, Gerardo; Salunke, Rahul; Mfarrej, Sara; Rachmadi, Andri Taruna; Hala, Sharif; Xu, Jinna; Alofi, Fadwa S; Khogeer, Asim; Hashem, Anwar M; Almontashiri, Naif AM; Alsomali, Afrah; Hamdan, Samir; Hong, Pei-Ying; Pain, Arnab; Li, Mo (Cold Spring Harbor Laboratory, 2020-06-29) [Preprint]
    <jats:p>Diagnosis and surveillance of emerging pathogens such as SARS-CoV-2 depend on nucleic acid isolation from clinical and environmental samples. Under normal circumstances, samples would be processed using commercial proprietary reagents in Biosafety 2 (BSL-2) or higher facilities. A pandemic at the scale of COVID-19 has caused a global shortage of proprietary reagents and BSL-2 laboratories to safely perform testing. Therefore, alternative solutions are urgently needed to address these challenges. We developed an open-source method called Magnetic- nanoparticle-Aided Viral RNA Isolation of Contagious Samples (MAVRICS) that is built upon reagents that are either readily available or can be synthesized in any molecular biology laboratory with basic equipment. Unlike conventional methods, MAVRICS works directly in samples inactivated in acid guanidinium thiocyanate-phenol-chloroform (e.g., TRIzol), thus allowing infectious samples to be handled safely without biocontainment facilities. Using 36 COVID-19 patient samples, 2 wastewater samples and 1 human pathogens control sample, we showed that MAVRICS rivals commercial kits in validated diagnostic tests of SARS-CoV-2, influenza viruses, and respiratory syncytial virus. MAVRICS is scalable and thus could become an enabling technology for widespread community testing and wastewater monitoring in the current and future pandemics.</jats:p>
  • A Nonmonotone Matrix-Free Algorithm for Nonlinear Equality-Constrained Inverse Problems

    Bergou, El Houcine; Diouane, Y.; Kungurtsev, V.; Royer, C. W. (arXiv, 2020-06-29) [Preprint]
    Least squares form one of the most prominent classes of optimization problems, with numerous applications in scientific computing and data fitting. When such formulations aim at modeling complex systems, the optimization process must account for nonlinear dynamics by incorporating constraints. In addition, these systems often incorporate a large number of variables, which increases the difficulty of the problem, and motivates the need for efficient algorithms amenable to large-scale implementations. In this paper, we propose and analyze a Levenberg-Marquardt algorithm for nonlinear least squares subject to nonlinear equality constraints. Our algorithm is based on inexact solves of linear least-squares problems, that only require Jacobian-vector products. Global convergence is guaranteed by the combination of a composite step approach and a nonmonotone step acceptance rule. We illustrate the performance of our method on several test cases from data assimilation and inverse problems: our algorithm is able to reach the vicinity of a solution from an arbitrary starting point, and can outperform the most natural alternatives for these classes of problems.
  • A Derivative Tracking Model for Wind Power Forecast Error

    Caballero, Renzo; Kebaier, Ahmed; Scavino, Marco; Tempone, Raúl (arXiv, 2020-06-29) [Preprint]
    Reliable wind power generation forecasting is crucial for applications such as the allocation of energy reserves, optimization for electricity price, and operation scheduling of conventional power plants. We propose a data-driven model based on parametric Stochastic Differential Equations (SDEs) to capture the real asymmetric dynamics of wind power forecast errors. Our SDE framework features time-derivative tracking of the forecast, time-varying mean-reversion parameter, and an improved state-dependent diffusion term. The methodology we developed allows the simulation of future wind power production paths and to obtain sharp empirical confidence bands. All the procedures are agnostic of the forecasting technology, and they enable comparisons between different forecast providers. We apply the model to historical Uruguayan wind power production data and forecasts between April and December 2019.
  • Patterns, drivers, and ecological implications of upwelling in coral reef habitats of the southern Red Sea

    De Carlo, Thomas Mario; Carvalho, Susana; Gajdzik, Laura; Hardenstine, Royale; Tanabe, Lyndsey K; Villalobos, Rodrigo; Berumen, Michael L. (Wiley, 2020-06-28) [Preprint]
    Coral reef ecosystems are highly sensitive to thermal anomalies, making them vulnerable to ongoing global warming. Yet, a variety of cooling mechanisms, such as upwelling, offer some respite to certain reefs. The Farasan Banks in the southern Red Sea is home to hundreds of coral reefs covering 16,000 km and experiences among the highest water temperatures of any coral-reef region despite exposure to summertime upwelling. We deployed an array of temperature loggers on coral reefs in the Farasan Banks, enabling us to evaluate the skill of satellite-based sea surface temperature (SST) products for capturing patterns of upwelling. Additionally, we used remote sensing products to investigate the physical drivers of upwelling, and to better understand how upwelling modulates summertime heat stress on coral communities. Our results show that various satellite SST products underestimate reef-water temperatures but differ in their ability to capture the spatial and temporal dynamics of upwelling. Monsoon winds from June to September drive the upwelling in the southern Red Sea via Ekman transport of surface waters off the shelf, and this process is ultimately controlled by the southwest Indian monsoon in the Arabian Sea. Further, the timing of the cessation of monsoon winds regulates the maximum water temperatures that are reached in September and October. In addition to describing the patterns and mechanisms of upwelling, our study sheds light on the broad ecological implications of this upwelling system, including modulation of coral bleaching events and effects on biodiversity, sea turtle reproduction, fish pelagic larval duration, and planktivore populations.
  • Valid model-free spatial prediction

    Mao, Huiying; Martin, Ryan; Reich, Brian (arXiv, 2020-06-28) [Preprint]
    Predicting the response at an unobserved location is a fundamental problem in spatial statistics. Given the difficulty in modeling spatial dependence, especially in non-stationary cases, model-based prediction intervals are at risk of misspecification bias that can negatively affect their validity. Here we present a new approach for model-free spatial prediction based on the {\em conformal prediction} machinery. Our key observation is that spatial data can be treated as exactly or approximately exchangeable in a wide range of settings. For example, when the spatial locations are deterministic, we prove that the response values are, in a certain sense, locally approximately exchangeable for a broad class of spatial processes, and we develop a local spatial conformal prediction algorithm that yields valid prediction intervals without model assumptions. Numerical examples with both real and simulated data confirm that the proposed conformal prediction intervals are valid and generally more efficient than existing model-based procedures across a range of non-stationary and non-Gaussian settings.
  • On Integrated Access and Backhaul Networks: Current Status and Potentials

    Madapatha, Charitha; Makki, Behrooz; Fang, Chao; Teyeb, Oumer; Dahlman, Erik; Alouini, Mohamed-Slim; Svensson, Tommy (arXiv, 2020-06-25) [Preprint]
    In this paper, we introduce and study the potentials and challenges of integrated access and backhaul (IAB) as one of the promising techniques for evolving 5G networks. We study IAB networks from different perspectives. We summarize the recent Rel-16 as well as the upcoming Rel-17 3GPP discussions on IAB, and highlight the main IAB-specific agreements on different protocol layers. Also, concentrating on millimeter wave-based communications, we evaluate the performance of IAB networks in both dense and suburban areas. Using a finite stochastic geometry model, with random distributions of IAB nodes as well as user equipments (UEs) in a finite region, we study the service coverage rate defined as the probability of the event that the UEs' minimum rate requirements are satisfied. We present comparisons between IAB and hybrid IAB/fiber-backhauled networks where a part or all of the small base stations are fiber-connected. Finally, we study the robustness of IAB networks to weather and various deployment conditions and verify their effects, such as blockage, tree foliage, rain as well as antenna height/gain on the coverage rate of IAB setups, as the key differences between the fiber-connected and IAB networks. As we show, IAB is an attractive approach to enable the network densification required by 5G and beyond.
  • High-Dimensional Quadratic Discriminant Analysis under Spiked Covariance Model

    Sifaou, Houssem; Kammoun, Abla; Alouini, Mohamed-Slim (arXiv, 2020-06-25) [Preprint]
    Quadratic discriminant analysis (QDA) is a widely used classification technique that generalizes the linear discriminant analysis (LDA) classifier to the case of distinct covariance matrices among classes. For the QDA classifier to yield high classification performance, an accurate estimation of the covariance matrices is required. Such a task becomes all the more challenging in high dimensional settings, wherein the number of observations is comparable with the feature dimension. A popular way to enhance the performance of QDA classifier under these circumstances is to regularize the covariance matrix, giving the name regularized QDA (R-QDA) to the corresponding classifier. In this work, we consider the case in which the population covariance matrix has a spiked covariance structure, a model that is often assumed in several applications. Building on the classical QDA, we propose a novel quadratic classification technique, the parameters of which are chosen such that the fisher-discriminant ratio is maximized. Numerical simulations show that the proposed classifier not only outperforms the classical R-QDA for both synthetic and real data but also requires lower computational complexity, making it suitable to high dimensional settings.
  • Break Point Detection for Functional Covariance

    Jiao, Shuhao; Frostig, Ron D.; Ombao, Hernando (arXiv, 2020-06-24) [Preprint]
    Many experiments record sequential trajectories that oscillate around zero. Such trajectories can be viewed as zero-mean functional data. When there are structural breaks (on the sequence of curves) in higher order moments, it is often difficult to spot these by mere visual inspection. Thus, we propose a detection and testing procedure to find the change-points in functional covariance. The method is fully functional in the sense that no dimension reduction is needed. We establish the asymptotic properties of the estimated change-point. The effectiveness of the proposed method is numerically validated in the simulation studies and an application to study structural changes in rat brain signals in a stroke experiment.
  • Weaker cooling by aerosols due to dust-pollution interactions

    Klingmüller, Klaus; Karydis, Vlassis A.; Bacer, Sara; Stenchikov, Georgiy L.; Lelieveld, Jos (Copernicus GmbH, 2020-06-23) [Preprint]
    <jats:p>Abstract. The interactions between aeolian dust and anthropogenic air pollution, notably chemical ageing of mineral dust and coagulation of dust and pollution particles, modify the atmospheric aerosol composition and burden. Since the aerosol particles can act as cloud condensation nuclei, this not only affects the radiative transfer directly via aerosol-radiation interactions, but also indirectly through cloud adjustments. We study both radiative effects using the global ECHAM/MESSy atmospheric chemistry-climate model (EMAC) which combines the Modular Earth Submodel System (MESSy) with the European Centre/Hamburg (ECHAM) climate model. Our simulations show that dust-pollution interactions reduce the cloud water path and hence the reflection of solar radiation. The associated climate warming outweighs the cooling which the dust-pollution interactions exert through the direct radiative effect. In total, this results in a net warming by dust-pollution interactions which moderates the negative global anthropogenic aerosol forcing at the top of the atmosphere by (0.2 ± 0.1) W m−2. </jats:p>
  • Signal Shaping for Non-Uniform Beamspace Modulated mmWave Hybrid MIMO Communications

    Guo, Shuaishuai; Zhang, Haixia; Zhang, Peng; Zhang, Shuping; Xu, Chengcheng; Alouini, Mohamed-Slim (arXiv, 2020-06-23) [Preprint]
    This paper investigates adaptive signal shaping methods for millimeter wave (mmWave) multiple-input multiple-output (MIMO) communications based on the maximizing the minimum Euclidean distance (MMED) criterion. In this work, we utilize the indices of analog precoders to carry information and optimize the symbol vector sets used for each analog precoder activation state. Specifically, we firstly propose a joint optimization based signal shaping (JOSS) approach, in which the symbol vector sets used for all analog precoder activation states are jointly optimized by solving a series of quadratically constrained quadratic programming (QCQP) problems. JOSS exhibits good performance, however, with a high computational complexity. To reduce the computational complexity, we then propose a full precoding based signal shaping (FPSS) method and a diagonal precoding based signal shaping (DPSS) method, where the full or diagonal digital precoders for all analog precoder activation states are optimized by solving two small-scale QCQP problems. Simulation results show that the proposed signal shaping methods can provide considerable performance gain in reliability in comparison with existing mmWave transmission solutions.
  • Modeling quantitative traits for COVID-19 case reports

    Queralt-Rosinach, Núria; Bello, Susan; Hoehndorf, Robert; Weiland, Claus; Rocca-Serra, Philippe; Schofield, Paul N. (Cold Spring Harbor Laboratory, 2020-06-21) [Preprint]
    <jats:p>Medical practitioners record the condition status of a patient through qualitative and quantitative observations. The measurement of vital signs and molecular parameters in the clinics gives a complementary description of abnormal phenotypes associated with the progression of a disease. The Clinical Measurement Ontology (CMO) is used to standardize annotations of these measurable traits. However, researchers have no way to describe how these quantitative traits relate to phenotype concepts in a machine-readable manner. Using the WHO clinical case report form standard for the COVID-19 pandemic, we modeled quantitative traits and developed OWL axioms to formally relate clinical measurement terms with anatomical, biomolecular entities and phenotypes annotated with the Uber-anatomy ontology (Uberon), Chemical Entities of Biological Interest (ChEBI) and the Phenotype and Trait Ontology (PATO) biomedical ontologies. The formal description of these relations allows interoperability between clinical and biological descriptions, and facilitates automated reasoning for analysis of patterns over quantitative and qualitative biomedical observations.</jats:p>
  • Conditional Normal Extreme-Value Copulas

    Krupskii, Pavel; Genton, Marc G. (arXiv, 2020-06-21) [Preprint]
    We propose a new class of extreme-value copulas which are extreme-value limits of conditional normal models. Conditional normal models are generalizations of conditional independence models, where the dependence among observed variables is modeled using one unobserved factor. Conditional on this factor, the distribution of these variables is given by the Gaussian copula. This structure allows one to build flexible and parsimonious models for data with complex dependence structures, such as data with spatial or temporal dependence. We study the extreme-value limits of these models and show some interesting special cases of the proposed class of copulas. We develop estimation methods for the proposed models and conduct a simulation study to assess the performance of these algorithms. Finally, we apply these copula models to analyze data on monthly wind maxima and stock return minima.
  • Network Moments: Extensions and Sparse-Smooth Attacks

    Alfadly, Modar; Bibi, Adel; Botero, Emilio; Al-Subaihi, Salman; Ghanem, Bernard (arXiv, 2020-06-21) [Preprint]
    The impressive performance of deep neural networks (DNNs) has immensely strengthened the line of research that aims at theoretically analyzing their effectiveness. This has incited research on the reaction of DNNs to noisy input, namely developing adversarial input attacks and strategies that lead to robust DNNs to these attacks. To that end, in this paper, we derive exact analytic expressions for the first and second moments (mean and variance) of a small piecewise linear (PL) network (Affine, ReLU, Affine) subject to Gaussian input. In particular, we generalize the second-moment expression of Bibi et al. to arbitrary input Gaussian distributions, dropping the zero-mean assumption. We show that the new variance expression can be efficiently approximated leading to much tighter variance estimates as compared to the preliminary results of Bibi et al. Moreover, we experimentally show that these expressions are tight under simple linearizations of deeper PL-DNNs, where we investigate the effect of the linearization sensitivity on the accuracy of the moment estimates. Lastly, we show that the derived expressions can be used to construct sparse and smooth Gaussian adversarial attacks (targeted and non-targeted) that tend to lead to perceptually feasible input attacks.
  • Performance Analysis and Optimization of Cooperative Satellite-Aerial-Terrestrial Systems

    Pan, Gaofeng; Ye, Jia; Zhang, Yongqiang; Alouini, Mohamed-Slim (arXiv, 2020-06-21) [Preprint]
    Aerial relays have been regarded as an alternative and promising solution to extend and improve satellite-terrestrial communications, as the probability of line-of-sight transmissions increases compared with adopting terrestrial relays. In this paper, a cooperative satellite-aerial-terrestrial system including a satellite transmitter (S), a group of terrestrial receivers (D), and an aerial relay (R) is considered. Specifically, considering the randomness of S and D and employing stochastic geometry, the coverage probability of R-D links in non-interference and interference scenarios is studied, and the outage performance of S-R link is investigated by deriving an approximated expression for the outage probability. Moreover, an optimization problem in terms of the transmit power and the transmission time over S-R and R-D links is formulated and solved to obtain the optimal end-to-end energy efficiency for the considered system. Finally, some numerical results are provided to validate our proposed analysis models, as well as to study the optimal energy efficiency performance of the considered system.
  • Unified Analysis of Stochastic Gradient Methods for Composite Convex and Smooth Optimization

    Khaled, Ahmed; Sebbouh, Othmane; Loizou, Nicolas; Gower, Robert M.; Richtarik, Peter (arXiv, 2020-06-20) [Preprint]
    We present a unified theorem for the convergence analysis of stochastic gradient algorithms for minimizing a smooth and convex loss plus a convex regularizer. We do this by extending the unified analysis of Gorbunov, Hanzely \& Richt\'arik (2020) and dropping the requirement that the loss function be strongly convex. Instead, we only rely on convexity of the loss function. Our unified analysis applies to a host of existing algorithms such as proximal SGD, variance reduced methods, quantization and some coordinate descent type methods. For the variance reduced methods, we recover the best known convergence rates as special cases. For proximal SGD, the quantization and coordinate type methods, we uncover new state-of-the-art convergence rates. Our analysis also includes any form of sampling and minibatching. As such, we are able to determine the minibatch size that optimizes the total complexity of variance reduced methods. We showcase this by obtaining a simple formula for the optimal minibatch size of two variance reduced methods (\textit{L-SVRG} and \textit{SAGA}). This optimal minibatch size not only improves the theoretical total complexity of the methods but also improves their convergence in practice, as we show in several experiments.
  • Cytosine deamination in SARS-CoV-2 leads to progressive CpG depletion.

    Sadykov, Mukhtar; Mourier, Tobias; Guan, Qingtian; Pain, Arnab (Cold Spring Harbor Laboratory, 2020-06-20) [Preprint]
    <jats:p>RNA viruses use CpG reduction to evade the host cell defense, but the driving mechanism is still largely unknown. To address this, we used rapidly growing genomic dataset of SARS-CoV-2 with relevant metadata information. SARS-CoV-2 genomes show a progressive increase of C-to-U substitutions resulting in CpG loss over just a few months. This is consistent with APOBEC-mediated RNA editing resulting in CpG reduction, thus allowing the virus to escape ZAP-mediated RNA degradation. Our results thus link the dynamics of target sequences in viral genome for two known host molecular defense mechanisms, the APOBEC and ZAP proteins.</jats:p>
  • Normalization Matters in Zero-Shot Learning

    Skorokhodov, Ivan; Elhoseiny, Mohamed (arXiv, 2020-06-19) [Preprint]
    An ability to grasp new concepts from their descriptions is one of the key features of human intelligence, and zero-shot learning (ZSL) aims to incorporate this property into machine learning models. In this paper, we theoretically investigate two very popular tricks used in ZSL: "normalize+scale" trick and attributes normalization and show how they help to preserve a signal's variance in a typical model during a forward pass. Next, we demonstrate that these two tricks are not enough to normalize a deep ZSL network. We derive a new initialization scheme, which allows us to demonstrate strong state-of-the-art results on 4 out of 5 commonly used ZSL datasets: SUN, CUB, AwA1, and AwA2 while being on average 2 orders faster than the closest runner-up. Finally, we generalize ZSL to a broader problem -- Continual Zero-Shot Learning (CZSL) and test our ideas in this new setup. The source code to reproduce all the results is available at https://github.com/universome/czsl.

View more