Showing papers by "Bar-Ilan University published in 2018"

Synthetic data augmentation using GAN for improved liver lesion classification

Abstract: In this paper, we present a data augmentation method that generates synthetic medical images using Generative Adversarial Networks (GANs). We propose a training scheme that first uses classical data augmentation to enlarge the training set and then further enlarges the data size and its diversity by applying GAN techniques for synthetic data augmentation. Our method is demonstrated on a limited dataset of computed tomography (CT) images of 182 liver lesions (53 cysts, 64 metastases and 65 hemangiomas). The classification performance using only classic data augmentation yielded 78.6% sensitivity and 88.4% specificity. By adding the synthetic data augmentation the results significantly increased to 85.7% sensitivity and 92.4% specificity.

On the Use of Unmanned Aerial Systems for Environmental Monitoring

Abstract: Environmental monitoring plays a central role in diagnosing climate and management impacts on natural and agricultural systems; enhancing the understanding of hydrological processes; optimizing the allocation and distribution of water resources; and assessing, forecasting, and even preventing natural disasters. Nowadays, most monitoring and data collection systems are based upon a combination of ground-based measurements, manned airborne sensors, and satellite observations. These data are utilized in describing both small- and large-scale processes, but have spatiotemporal constraints inherent to each respective collection system. Bridging the unique spatial and temporal divides that limit current monitoring platforms is key to improving our understanding of environmental systems. In this context, Unmanned Aerial Systems (UAS) have considerable potential to radically improve environmental monitoring. UAS-mounted sensors offer an extraordinary opportunity to bridge the existing gap between field observations and traditional air- and space-borne remote sensing, by providing high spatial detail over relatively large areas in a cost-effective way and an entirely new capacity for enhanced temporal retrieval. As well as showcasing recent advances in the field, there is also a need to identify and understand the potential limitations of UAS technology. For these platforms to reach their monitoring potential, a wide spectrum of unresolved issues and application-specific challenges require focused community attention. Indeed, to leverage the full potential of UAS-based approaches, sensing technologies, measurement protocols, postprocessing techniques, retrieval algorithms, and evaluation techniques need to be harmonized. The aim of this paper is to provide an overview of the existing research and applications of UAS in natural and agricultural ecosystem monitoring in order to identify future directions, applications, developments, and challenges.

From Surface ZrO2 Coating to Bulk Zr Doping by High Temperature Annealing of Nickel-Rich Lithiated Oxides and Their Enhanced Electrochemical Performance in Lithium Ion Batteries

Abstract: One of the major hurdles of Ni-rich cathode materials Li1+x(NixCozMnz)wO2, y > 0.5 for lithium-ion batteries is their low cycling stability especially for compositions with Ni ≥ 60%, which suffer from severe capacity fading and impedance increase during cycling at elevated temperatures (e.g., 45 °C). Two promising surface and structural modifications of these materials to alleviate the above drawback are (1) coatings by electrochemically inert inorganic compounds (e.g., ZrO2) or (2) lattice doping by cations like Zr4+, Al3+, Mg2+, etc. This paper demonstrates the enhanced electrochemical behavior of Ni-rich material LiNi0.8Co0.1Mn0.1O2 (NCM811) coated with a thin ZrO2 layer. The coating is produced by an easy and scalable wet chemical approach followed by annealing the material at ≥700 °C under oxygen that results in Zr doping. It is established that some ZrO2 remains even after annealing at ≥800 °C as a surface layer on NCM811. The main finding of this work is the enhanced cycling stability and lower impedance of the coated/doped NCM811 that can be attributed to a synergetic effect of the ZrO2 coating in combination with a zirconium doping.

Breaking NLI Systems with Sentences that Require Simple Lexical Inferences

Abstract: We create a new NLI test set that shows the deficiency of state-of-the-art models in inferences that require lexical and world knowledge. The new examples are simpler than the SNLI test set, containing sentences that differ by at most one word from sentences in the training set. Yet, the performance on the new test set is substantially worse across systems trained on SNLI, demonstrating that these systems are limited in their generalization ability, failing to capture many simple inferences.

Large-scale replicated field study of maize rhizosphere identifies heritable microbes

Abstract: Soil microbes that colonize plant roots and are responsive to differences in plant genotype remain to be ascertained for agronomically important crops. From a very large-scale longitudinal field study of 27 maize inbred lines planted in three fields, with partial replication 5 y later, we identify root-associated microbiota exhibiting reproducible associations with plant genotype. Analysis of 4,866 samples identified 143 operational taxonomic units (OTUs) whose variation in relative abundances across the samples was significantly regulated by plant genotype, and included five of seven core OTUs present in all samples. Plant genetic effects were significant amid the large effects of plant age on the rhizosphere microbiome, regardless of the specific community of each field, and despite microbiome responses to climate events. Seasonal patterns showed that the plant root microbiome is locally seeded, changes with plant growth, and responds to weather events. However, against this background of variation, specific taxa responded to differences in host genotype. If shown to have beneficial functions, microbes may be considered candidate traits for selective breeding.

A-to-I RNA editing — immune protector and transcriptome diversifier

Abstract: Modifications of RNA affect its function and stability. RNA editing is unique among these modifications because it not only alters the cellular fate of RNA molecules but also alters their sequence relative to the genome. The most common type of RNA editing is A-to-I editing by double-stranded RNA-specific adenosine deaminase (ADAR) enzymes. Recent transcriptomic studies have identified a number of 'recoding' sites at which A-to-I editing results in non-synonymous substitutions in protein-coding sequences. Many of these recoding sites are conserved within (but not usually across) lineages, are under positive selection and have functional and evolutionary importance. However, systematic mapping of the editome across the animal kingdom has revealed that most A-to-I editing sites are located within mobile elements in non-coding parts of the genome. Editing of these non-coding sites is thought to have a critical role in protecting against activation of innate immunity by self-transcripts. Both recoding and non-coding events have implications for genome evolution and, when deregulated, may lead to disease. Finally, ADARs are now being adapted for RNA engineering purposes.

Pushing the limit of layered transition metal oxide cathodes for high-energy density rechargeable Li ion batteries

Abstract: Development of advanced high energy density lithium ion batteries is important for promoting electromobility. Making electric vehicles attractive and competitive compared to conventional automobiles depends on the availability of reliable, safe, high power, and highly energetic batteries whose components are abundant and cost effective. Nickel rich Li[NixCoyMn1−x−y]O2 layered cathode materials (x > 0.5) are of interest because they can provide very high specific capacity without pushing charging potentials to levels that oxidize the electrolyte solutions. However, these cathode materials suffer from stability problems. We discovered that doping these materials with tungsten (1 mol%) remarkably increases their stability due to a partial layered to cubic (rock salt) phase transition. We demonstrate herein highly stable Li ion battery prototypes consisting of tungsten-stabilized Ni rich cathode materials (x > 0.9) with specific capacities >220 mA h g-1. This development can increase the energy density of Li ion batteries more than 30% above the state of the art without compromising durability.

Supervised Open Information Extraction

Abstract: We present data and methods that enable a supervised learning approach to Open Information Extraction (Open IE). Central to the approach is a novel formulation of Open IE as a sequence tagging problem, addressing challenges such as encoding multiple extractions for a predicate. We also develop a bi-LSTM transducer, extending recent deep Semantic Role Labeling models to extract Open IE tuples and provide confidence scores for tuning their precision-recall tradeoff. Furthermore, we show that the recently released Question-Answer Meaning Representation dataset can be automatically converted into an Open IE corpus which significantly increases the amount of available training data. Our supervised model outperforms the existing state-of-the-art Open IE systems on benchmark datasets.

Redox Mediators for Li-O2 Batteries: Status and Perspectives.

Abstract: Li-O2 batteries have received much attention due to their extremely large theoretical energy density. However, the high overpotentials required for charging Li-O2 batteries lower their energy efficiency and degrade the electrolytes and carbon electrodes. This problem is one of the main obstacles in developing practical Li-O2 batteries. To solve this problem, it is important to facilitate the oxidation of Li2 O2 upon charging by using effective electrocatalysis. Using solid catalysts is not too effective for oxidizing the electronically isolating Li-peroxide layers. In turn, for soluble catalysts, red-ox mediators (RMs) are homogeneously dissolved in the electrolyte solutions and can effectively oxidize all of the Li2 O2 precipitated during discharge. RMs can decompose solid Li2 O2 species no matter their size, morphology, or thickness and thus dramatically increase energy efficiency. However, some negative side effects, such as the shuttle reactions of RMs and deterioration of the Li-metal occur. Therefore, it is necessary to study the activity and stability of RMs in Li-O2 batteries in detail. Herein, recent studies related to redox mediators are reviewed and the mechanisms of redox reactions are illustrated. The development opportunities of RMs for this important battery technology are discussed and future directions are suggested.

The role of mitochondria in aging

Abstract: The biological basis of human aging remains one of the greatest unanswered scientific questions. Increasing evidence, however, points to a role for alterations in mitochondrial function as a potential central regulator of the aging process. Here, we focus primarily on three aspects of mitochondrial biology that link this ancient organelle to how and why we age. In particular, we discuss the role of mitochondria in regulating the innate immune system, the mechanisms linking mitochondrial quality control to age-dependent pathology, and the possibility that mitochondrial-to-nuclear signaling might regulate the rate of aging.

Audio source separation and speech enhancement

Abstract: Audio source separation and speech enhancement aim to extract one or more source signals of interest from an audio recording involving several sound sources. These technologies are among the most studied in audio signal processing today and bear a critical role in the success of hearing aids, hands-free phones, voice command and other noise-robust audio analysis systems, and music post-production software.

Life-Course Genome-wide Association Study Meta-analysis of Total Body BMD and Assessment of Age-Specific Effects

Abstract: Bone mineral density (BMD) assessed by DXA is used to evaluate bone health. In children, total body (TB) measurements are commonly used; in older individuals, BMD at the lumbar spine (LS) and femoral neck (FN) is used to diagnose osteoporosis. To date, genetic variants in more than 60 loci have been identified as associated with BMD. To investigate the genetic determinants of TB-BMD variation along the life course and test for age-specific effects, we performed a meta-analysis of 30 genome-wide association studies (GWASs) of TB-BMD including 66,628 individuals overall and divided across five age strata, each spanning 15 years. We identified variants associated with TB-BMD at 80 loci, of which 36 have not been previously identified; overall, they explain approximately 10% of the TB-BMD variance when combining all age groups and influence the risk of fracture. Pathway and enrichment analysis of the association signals showed clustering within gene sets implicated in the regulation of cell growth and SMAD proteins, overexpressed in the musculoskeletal system, and enriched in enhancer and promoter regions. These findings reveal TB-BMD as a relevant trait for genetic studies of osteoporosis, enabling the identification of variants and pathways influencing different bone compartments. Only variants in ESR1 and close proximity to RANKL showed a clear effect dependency on age. This most likely indicates that the majority of genetic variants identified influence BMD early in life and that their effect can be captured throughout the life course.

Adversarial Removal of Demographic Attributes from Text Data

Abstract: Recent advances in Representation Learning and Adversarial Training seem to succeed in removing unwanted features from the learned representation. We show that demographic information of authors is encoded in—and can be recovered from—the intermediate representations learned by text-based neural classifiers. The implication is that decisions of classifiers trained on textual data are not agnostic to—and likely condition on—demographic attributes. When attempting to remove such demographic information using adversarial training, we find that while the adversarial component achieves chance-level development-set accuracy during training, a post-hoc classifier, trained on the encoded sentences from the first part, still manages to reach substantially higher classification accuracies on the same data. This behavior is consistent across several tasks, demographic properties and datasets. We explore several techniques to improve the effectiveness of the adversarial component. Our main conclusion is a cautionary one: do not rely on the adversarial training to achieve invariant representation to sensitive features.

Understanding Convolutional Neural Networks for Text Classification

Abstract: We present an analysis into the inner workings of Convolutional Neural Networks (CNNs) for processing text. CNNs used for computer vision can be interpreted by projecting filters into image space, but for discrete sequence inputs CNNs remain a mystery. We aim to understand the method by which the networks process and classify text. We examine common hypotheses to this problem: that filters, accompanied by global max-pooling, serve as ngram detectors. We show that filters may capture several different semantic classes of ngrams by using different activation patterns, and that global max-pooling induces behavior which separates important ngrams from the rest. Finally, we show practical use cases derived from our findings in the form of model interpretability (explaining a trained model by deriving a concrete identity for each filter, bridging the gap between visualization tools in vision tasks and NLP) and prediction interpretability (explaining predictions).

Radiation induces proinflammatory dysbiosis: transmission of inflammatory susceptibility by host cytokine induction

Abstract: Objective Radiation proctitis (RP) is a complication of pelvic radiotherapy which affects both the host and microbiota. Herein we assessed the radiation effect on microbiota and its relationship to tissue damage using a rectal radiation mouse model. Design We evaluated luminal and mucosa-associated dysbiosis in irradiated and control mice at two postradiation time points and correlated it with clinical and immunological parameters. Epithelial cytokine response was evaluated using bacterial–epithelial co-cultures. Subsequently, germ-free (GF) mice were colonised with postradiation microbiota and controls and exposed to radiation, or dextran sulfate-sodium (DSS). Interleukin (IL)-1β correlated with tissue damage and was induced by dysbiosis. Therefore, we tested its direct role in radiation-induced damage by IL-1 receptor antagonist administration to irradiated mice. Results A postradiation shift in microbiota was observed. A unique microbial signature correlated with histopathology. Increased colonic tumor necrosis factor (TNF)α, IL-1β and IL-6 expression was observed at two different time points. Adherent microbiota from RP differed from those in uninvolved segments and was associated with tissue damage. Using bacterial–epithelial co-cultures, postradiation microbiota enhanced IL-1β and TNFα expression compared with naive microbiota. GF mice colonisation by irradiated microbiota versus controls predisposed mice to both radiation injury and DSS-induced colitis. IL-1 receptor antagonist administration ameliorated intestinal radiation injury. Conclusions The results demonstrate that rectal radiation induces dysbiosis, which transmits radiation and inflammatory susceptibility and provide evidence that microbial-induced radiation tissue damage is at least in part mediated by IL-1β. Environmental factors may affect the host via modifications of the microbiome and potentially allow for novel interventional approaches via its manipulation.

Glucose-Responsive Metal–Organic-Framework Nanoparticles Act as “Smart” Sense-and-Treat Carriers

Abstract: Zeolitic Zn2+-imidazolate cross-linked framework nanoparticles, ZIF-8 NMOFs, are used as “smart” glucose-responsive carriers for the controlled release of drugs. The ZIF-8 NMOFs are loaded with the respective drug and glucose oxidase (GOx), and the GOx-mediated aerobic oxidation of glucose yields gluconic acid and H2O2. The acidification of the NMOFs’ microenvironment leads to the degradation of the nanoparticles and the release of the loaded drugs. In one sense-and-treat system, GOx and insulin are loaded in the NMOFs. In the presence of glucose, the nanoparticles are unlocked, resulting in the release of insulin. The release of insulin is controlled by the concentration of glucose. In the second sense-and-treat system, the NMOFs are loaded with the antivascular endothelial growth factor aptamer (VEGF aptamer) and GOx. In the presence of glucose, the ZIF-8 NMOFs are degraded, leading to the release of the VEGF aptamer, which acts as a potential inhibitor of the angiogenetic regeneration of blood vessels ...

SBFT: a Scalable and Decentralized Trust Infrastructure

Abstract: SBFT is a state of the art Byzantine fault tolerant permissioned blockchain system that addresses the challenges of scalability, decentralization and world-scale geo-replication. SBFTis optimized for decentralization and can easily handle more than 200 active replicas in a real world-scale deployment. We evaluate \sysname in a world-scale geo-replicated deployment with 209 replicas withstanding f=64 Byzantine failures. We provide experiments that show how the different algorithmic ingredients of \sysname increase its performance and scalability. The results show that SBFT simultaneously provides almost 2x better throughput and about 1.5x better latency relative to a highly optimized system that implements the PBFT protocol. To achieve this performance improvement, SBFT uses a combination of four ingredients: using collectors and threshold signatures to reduce communication to linear, using an optimistic fast path, reducing client communication and utilizing redundant servers for the fast path.

Turning your weakness into a strength: watermarking deep neural networks by backdooring

Abstract: Deep Neural Networks have recently gained lots of success after enabling several breakthroughs in notoriously challenging problems. Training these networks is computationally expensive and requires vast amounts of training data. Selling such pre-trained models can, therefore, be a lucrative business model. Unfortunately, once the models are sold they can be easily copied and redistributed. To avoid this, a tracking mechanism to identify models as the intellectual property of a particular vendor is necessary. In this work, we present an approach for watermarking Deep Neural Networks in a black-box way. Our scheme works for general classification tasks and can easily be combined with current learning algorithms. We show experimentally that such a watermark has no noticeable impact on the primary task that the model is designed for and evaluate the robustness of our proposal against a multitude of practical attacks. Moreover, we provide a theoretical analysis, relating our approach to previous work on backdooring.

Prognostic implications of atrial fibrillation in heart failure with reduced, mid-range, and preserved ejection fraction: a report from 14 964 patients in the European Society of Cardiology Heart Failure Long-Term Registry

Abstract: Aim To investigate the characteristics long-term prognostic implications (up to ∼2.2 years) of atrial fibrillation (AF) compared to sinus rhythm (SR), between acute and chronic heart failure (HF) with reduced (HFrEF < 40%), mid-range (HFmrEF 40-49%), and preserved (HFpEF ≥ 50%) ejection fraction (EF). Methods and results Data from the observational, prospective, HF long-term registry of the European Society of Cardiology were analysed. A total of 14 964 HF patients (age 66 ± 13 years, 67% male; 53% HFrEF, 21% HFmrEF, 26% HFpEF) were enrolled. The prevalence of AF was 27% in HFrEF, 29% in HFmrEF, and 39% in HFpEF. Atrial fibrillation was associated with older age, lower functional capacity, and heightened physical signs of HF. Crude rates of mortality and HF hospitalization were higher in patients with AF compared to SR, in each EF subtype. After multivariable adjustment, the hazard ratio of AF for HF hospitalizations was: 1.036 (95% CI 0.888-1.208, P = 0.652) in HFrEF, 1.430 (95% CI 1.087-1.882, P = 0.011) in HFmrEF, and 1.487 (95% CI 1.195-1.851, P < 0.001) in HFpEF; and for combined all-cause death or HF hospitalizations: 0.957 (95% CI 0.843-1.087, P = 0.502), 1.302 (95% CI 1.055-1.608, P = 0.014), and 1.365 (95% CI 1.152-1.619, P < 0.001), respectively. In patients with HFrEF, AF was not associated with worse outcomes in those presenting with either an acute or a chronic presentation of HF. Conclusions The prevalence of AF increases with increasing EF but its association with worse cardiovascular outcomes, remained significant in patients with HFpEF and HFmrEF, but not in those with HFrEF.

On the Practical Computational Power of Finite Precision RNNs for Language Recognition

Abstract: While Recurrent Neural Networks (RNNs) are famously known to be Turing complete, this relies on infinite precision in the states and unbounded computation time. We consider the case of RNNs with finite precision whose computation time is linear in the input length. Under these limitations, we show that different RNN variants have different computational power. In particular, we show that the LSTM and the Elman-RNN with ReLU activation are strictly stronger than the RNN with a squashing activation and the GRU. This is achieved because LSTMs and ReLU-RNNs can easily implement counting behavior. We show empirically that the LSTM does indeed learn to effectively use the counting mechanism.

Antibiotics in early life: dysbiosis and the damage done.

Abstract: Antibiotics are the most common type of medication prescribed to children, including infants, in the Western world. While use of antibiotics has transformed previously lethal infections into relatively minor diseases, antibiotic treatments can have adverse effects as well. It has been shown in children, adults and animal models that antibiotics dramatically alter the gut microbial composition. Since the gut microbiota plays crucial roles in immunity, metabolism and endocrinology, the effects of antibiotics on the microbiota may lead to further health complications. In this review, we present an overview of the effects of antibiotics on the microbiome in children, and correlate them to long-lasting complications of obesity, behavior, allergies, autoimmunity and other diseases.

Scalable Private Set Intersection Based on OT Extension

Abstract: Private set intersection (PSI) allows two parties to compute the intersection of their sets without revealing any information about items that are not in the intersection. It is one of the best studied applications of secure computation and many PSI protocols have been proposed. However, the variety of existing PSI protocols makes it difficult to identify the solution that performs best in a respective scenario, especially since they were not compared in the same setting. In addition, existing PSI protocols are several orders of magnitude slower than an insecure naive hashing solution, which is used in practice.In this article, we review the progress made on PSI protocols and give an overview of existing protocols in various security models. We then focus on PSI protocols that are secure against semi-honest adversaries and take advantage of the most recent efficiency improvements in Oblivious Transfer (OT) extension, propose significant optimizations to previous PSI protocols, and suggest a new PSI protocol whose runtime is superior to that of existing protocols. We compare the performance of the protocols, both theoretically and experimentally, by implementing all protocols on the same platform, give recommendations on which protocol to use in a particular setting, and evaluate the progress on PSI protocols by comparing them to the currently employed insecure naive hashing protocol. We demonstrate the feasibility of our new PSI protocol by processing two sets with a billion elements each.

A parcel locker network as a solution to the logistics last mile problem

Abstract: We consider the problem of designing a parcel locker network as a solution to the Logistics Last Mile Problem: Choosing the optimal number, locations, and sizes of parcel locekers facilities. The o...

Fooling End-To-End Speaker Verification With Adversarial Examples

Abstract: Automatic speaker verification systems are increasingly used as the primary means to authenticate costumers. Recently, it has been proposed to train speaker verification systems using end-to-end deep neural models. In this paper, we show that such systems are vulnerable to adversarial example attacks. Adversarial examples are generated by adding a peculiar noise to original speaker examples, in such a way that they are almost indistinguishable, by a human listener. Yet, the generated waveforms, which sound as speaker A can be used to fool such a system by claiming as if the waveforms were uttered by speaker B. We present white-box attacks on a deep end-to-end network that was either trained on YOHO or NTIMIT. We also present two black-box attacks. In the first one, we generate adversarial examples with a system trained on NTIMIT and perform the attack on a system that trained on YOHO. In the second one, we generate the adversarial examples with a system trained using Mel-spectrum features and perform the attack on a system trained using MFCCs. Our results show that one can significantly decrease the accuracy of a target system even when the adversarial examples are generated with different system potentially using different features.

Breaking NLI Systems with Sentences that Require Simple Lexical Inferences

Abstract: We create a new NLI test set that shows the deficiency of state-of-the-art models in inferences that require lexical and world knowledge. The new examples are simpler than the SNLI test set, containing sentences that differ by at most one word from sentences in the training set. Yet, the performance on the new test set is substantially worse across systems trained on SNLI, demonstrating that these systems are limited in their generalization ability, failing to capture many simple inferences.