Showing papers by "École Normale Supérieure published in 2016"

DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks

Abstract: State-of-the-art deep neural networks have achieved impressive results on many image classification tasks. However, these same architectures have been shown to be unstable to small, well sought, perturbations of the images. Despite the importance of this phenomenon, no effective methods have been proposed to accurately compute the robustness of state-of-the-art deep classifiers to such perturbations on large-scale datasets. In this paper, we fill this gap and propose the DeepFool algorithm to efficiently compute perturbations that fool deep networks, and thus reliably quantify the robustness of these classifiers. Extensive experimental results show that our approach outperforms recent methods in the task of computing adversarial perturbations and making classifiers more robust.1

NetVLAD: CNN Architecture for Weakly Supervised Place Recognition

Abstract: We tackle the problem of large scale visual place recognition, where the task is to quickly and accurately recognize the location of a given query photograph. We present the following three principal contributions. First, we develop a convolutional neural network (CNN) architecture that is trainable in an end-to-end manner directly for the place recognition task. The main component of this architecture, NetVLAD, is a new generalized VLAD layer, inspired by the "Vector of Locally Aggregated Descriptors" image representation commonly used in image retrieval. The layer is readily pluggable into any CNN architecture and amenable to training via backpropagation. Second, we develop a training procedure, based on a new weakly supervised ranking loss, to learn parameters of the architecture in an end-to-end manner from images depicting the same places over time downloaded from Google Street View Time Machine. Finally, we show that the proposed architecture significantly outperforms non-learnt image representations and off-the-shelf CNN descriptors on two challenging place recognition benchmarks, and improves over current state of-the-art compact image representations on standard image retrieval benchmarks.

The rise of plastic bioelectronics

Abstract: Plastic bioelectronics is a research field that takes advantage of the inherent properties of polymers and soft organic electronics for applications at the interface of biology and electronics. The resulting electronic materials and devices are soft, stretchable and mechanically conformable, which are important qualities for interacting with biological systems in both wearable and implantable devices. Work is currently aimed at improving these devices with a view to making the electronic-biological interface as seamless as possible.

Transport in Out-of-Equilibrium XXZ Chains: Exact Profiles of Charges and Currents.

Abstract: We consider the nonequilibrium time evolution of piecewise homogeneous states in the XXZ spin-1/2 chain, a paradigmatic example of an interacting integrable model. The initial state can be thought of as the result of joining chains with different global properties. Through dephasing, at late times, the state becomes locally equivalent to a stationary state which explicitly depends on position and time. We propose a kinetic theory of elementary excitations and derive a continuity equation which fully characterizes the thermodynamics of the model. We restrict ourselves to the gapless phase and consider cases where the chains are prepared: (1) at different temperatures, (2) in the ground state of two different models, and (3) in the "domain wall" state. We find excellent agreement (any discrepancy is within the numerical error) between theoretical predictions and numerical simulations of time evolution based on time-evolving block decimation algorithms. As a corollary, we unveil an exact expression for the expectation values of the charge currents in a generic stationary state.

Structural control of mixed ionic and electronic transport in conducting polymers.

Abstract: Poly(3,4-ethylenedioxythiophene) doped with poly(styrenesulfonate), PEDOT:PSS, has been utilized for over two decades as a stable, solution-processable hole conductor. While its hole transport properties have been the subject of intense investigation, recent work has turned to PEDOT:PSS as a mixed ionic/electronic conductor in applications including bioelectronics, energy storage and management, and soft robotics. Conducting polymers can efficiently transport both holes and ions when sufficiently hydrated, however, little is known about the role of morphology on mixed conduction. Here, we show that bulk ionic and electronic mobilities are simultaneously affected by processing-induced changes in nano- and meso-scale structure in PEDOT:PSS films. We quantify domain composition, and find that domain purification on addition of dispersion co-solvents limits ion mobility, even while electronic conductivity improves. We show that an optimal morphology allows for the balanced ionic and electronic transport that is critical for prototypical mixed conductor devices. These findings may pave the way for the rational design of polymeric materials and processing routes to enhance devices reliant on mixed conduction. Conducting polymers are promising materials for applications including bioelectronics and soft robotics, but little is known about how morphology affects mixed conduction. Here, the authors show how bulk ionic/electronic transport is affected by changes in nano- and meso-scale structure in PEDOT:PSS films.

Ecogenomics and potential biogeochemical impacts of globally abundant ocean viruses

Abstract: Ocean microbes drive biogeochemical cycling on a global scale. However, this cycling is constrained by viruses that affect community composition, metabolic activity, and evolutionary trajectories. Owing to challenges with the sampling and cultivation of viruses, genome-level viral diversity remains poorly described and grossly understudied, with less than 1% of observed surface-ocean viruses known. Here we assemble complete genomes and large genomic fragments from both surface- and deep-ocean viruses sampled during the Tara Oceans and Malaspina research expeditions, and analyse the resulting 'global ocean virome' dataset to present a global map of abundant, double-stranded DNA viruses complete with genomic and ecological contexts. A total of 15,222 epipelagic and mesopelagic viral populations were identified, comprising 867 viral clusters (defined as approximately genus-level groups). This roughly triples the number of known ocean viral populations and doubles the number of candidate bacterial and archaeal virus genera, providing a near-complete sampling of epipelagic communities at both the population and viral-cluster level. We found that 38 of the 867 viral clusters were locally or globally abundant, together accounting for nearly half of the viral populations in any global ocean virome sample. While two-thirds of these clusters represent newly described viruses lacking any cultivated representative, most could be computationally linked to dominant, ecologically relevant microbial hosts. Moreover, we identified 243 viral-encoded auxiliary metabolic genes, of which only 95 were previously known. Deeper analyses of four of these auxiliary metabolic genes (dsrC, soxYZ, P-II (also known as glnB) and amoC) revealed that abundant viruses may directly manipulate sulfur and nitrogen cycling throughout the epipelagic ocean. This viral catalog and functional analyses provide a necessary foundation for the meaningful integration of viruses into ecosystem models where they act as key players in nutrient cycling and trophic networks.

Nonclassical Routes for Amide Bond Formation

Abstract: The present review offers an overview of nonclassical (e.g., with no pre- or in situ activation of a carboxylic acid partner) approaches for the construction of amide bonds. The review aims to comprehensively discuss relevant work, which was mainly done in the field in the last 20 years. Organization of the data follows a subdivision according to substrate classes: catalytic direct formation of amides from carboxylic and amines (section 2); the use of carboxylic acid surrogates (section 3); and the use of amine surrogates (section 4). The ligation strategies (NCL, Staudinger, KAHA, KATs, etc.) that could involve both carboxylic acid and amine surrogates are treated separately in section 5.

Plankton networks driving carbon export in the oligotrophic ocean.

Abstract: The biological carbon pump is the process by which CO2 is transformed to organic carbon via photosynthesis, exported through sinking particles, and finally sequestered in the deep ocean. While the intensity of the pump correlates with plankton community composition, the underlying ecosystem structure driving the process remains largely uncharacterized. Here we use environmental and metagenomic data gathered during the Tara Oceans expedition to improve our understanding of carbon export in the oligotrophic ocean. We show that specific plankton communities, from the surface and deep chlorophyll maximum, correlate with carbon export at 150 m and highlight unexpected taxa such as Radiolaria and alveolate parasites, as well as Synechococcus and their phages, as lineages most strongly associated with carbon export in the subtropical, nutrient-depleted, oligotrophic ocean. Additionally, we show that the relative abundance of a few bacterial and viral genes can predict a significant fraction of the variability in carbon export in these regions.

Identification of an anti-inflammatory protein from Faecalibacterium prausnitzii, a commensal bacterium deficient in Crohn’s disease

Abstract: Background Crohn’s disease (CD)-associated dysbiosis is characterised by a loss of Faecalibacterium prausnitzii, whose culture supernatant exerts an anti-inflammatory effect both in vitro and in vivo. However, the chemical nature of the anti-inflammatory compounds has not yet been determined. Methods Peptidomic analysis using mass spectrometry was applied to F. prausnitzii supernatant. Anti-inflammatory effects of identified peptides were tested in vitro directly on intestinal epithelial cell lines and on cell lines transfected with a plasmid construction coding for the candidate protein encompassing these peptides. In vivo, the cDNA of the candidate protein was delivered to the gut by recombinant lactic acid bacteria to prevent dinitrobenzene sulfonic acid (DNBS)-colitis in mice. Results The seven peptides, identified in the F. prausnitzii culture supernatants, derived from a single microbial anti-inflammatory molecule (MAM), a protein of 15 kDa, and comprising 53% of non-polar residues. This last feature prevented the direct characterisation of the putative anti-inflammatory activity of MAM-derived peptides. Transfection of MAM cDNA in epithelial cells led to a significant decrease in the activation of the nuclear factor (NF)-κB pathway with a dose-dependent effect. Finally, the use of a food-grade bacterium, Lactococcus lactis, delivering a plasmid encoding MAM was able to alleviate DNBS-induced colitis in mice. Conclusions A 15 kDa protein with anti-inflammatory properties is produced by F. prausnitzii, a commensal bacterium involved in CD pathogenesis. This protein is able to inhibit the NF-κB pathway in intestinal epithelial cells and to prevent colitis in an animal model.

Massive radius-dependent flow slippage in carbon nanotubes

Abstract: Measurements and simulations have found that water moves through carbon nanotubes at exceptionally high rates owing to nearly frictionless interfaces. These observations have stimulated interest in nanotube-based membranes for applications including desalination, nano-filtration and energy harvesting, yet the exact mechanisms of water transport inside the nanotubes and at the water-carbon interface continue to be debated because existing theories do not provide a satisfactory explanation for the limited number of experimental results available so far. This lack of experimental results arises because, even though controlled and systematic studies have explored transport through individual nanotubes, none has met the considerable technical challenge of unambiguously measuring the permeability of a single nanotube. Here we show that the pressure-driven flow rate through individual nanotubes can be determined with unprecedented sensitivity and without dyes from the hydrodynamics of water jets as they emerge from single nanotubes into a surrounding fluid. Our measurements reveal unexpectedly large and radius-dependent surface slippage in carbon nanotubes, and no slippage in boron nitride nanotubes that are crystallographically similar to carbon nanotubes, but electronically different. This pronounced contrast between the two systems must originate from subtle differences in the atomic-scale details of their solid-liquid interfaces, illustrating that nanofluidics is the frontier at which the continuum picture of fluid mechanics meets the atomic nature of matter.

Understanding deep convolutional networks.

Abstract: Deep convolutional networks provide state-of-the-art classifications and regressions results over many high-dimensional problems. We review their architecture, which scatters data with a cascade of linear filter weights and nonlinearities. A mathematical framework is introduced to analyse their properties. Computations of invariants involve multiscale contractions with wavelets, the linearization of hierarchical symmetries and sparse separations. Applications are discussed.

Insights into global diatom distribution and diversity in the world’s ocean

Abstract: Diatoms (Bacillariophyta) constitute one of the most diverse and ecologically important groups of phytoplankton. They are considered to be particularly important in nutrient-rich coastal ecosystems and at high latitudes, but considerably less so in the oligotrophic open ocean. The Tara Oceans circumnavigation collected samples from a wide range of oceanic regions using a standardized sampling procedure. Here, a total of ∼12 million diatom V9-18S ribosomal DNA (rDNA) ribotypes, derived from 293 size-fractionated plankton communities collected at 46 sampling sites across the global ocean euphotic zone, have been analyzed to explore diatom global diversity and community composition. We provide a new estimate of diversity of marine planktonic diatoms at 4,748 operational taxonomic units (OTUs). Based on the total assigned ribotypes, Chaetoceros was the most abundant and diverse genus, followed by Fragilariopsis, Thalassiosira, and Corethron. We found only a few cosmopolitan ribotypes displaying an even distribution across stations and high abundance, many of which could not be assigned with confidence to any known genus. Three distinct communities from South Pacific, Mediterranean, and Southern Ocean waters were identified that share a substantial percentage of ribotypes within them. Sudden drops in diversity were observed at Cape Agulhas, which separates the Indian and Atlantic Oceans, and across the Drake Passage between the Atlantic and Southern Oceans, indicating the importance of these ocean circulation choke points in constraining diatom distribution and diversity. We also observed high diatom diversity in the open ocean, suggesting that diatoms may be more relevant in these oceanic systems than generally considered.

Efficient Monolithic Perovskite/Silicon Tandem Solar Cell with Cell Area >1 cm2

Abstract: Monolithic perovskite/crystalline silicon tandem solar cells hold great promise for further performance improvement of well-established silicon photovoltaics; however, monolithic tandem integration is challenging, evidenced by the modest performances and small-area devices reported so far. Here we present first a low-temperature process for semitransparent perovskite solar cells, yielding efficiencies of up to 14.5%. Then, we implement this process to fabricate monolithic perovskite/silicon heterojunction tandem solar cells yielding efficiencies of up to 21.2 and 19.2% for cell areas of 0.17 and 1.22 cm2, respectively. Both efficiencies are well above those of the involved subcells. These single-junction perovskite and tandem solar cells are hysteresis-free and demonstrate steady performance under maximum power point tracking for several minutes. Finally, we present the effects of varying the intermediate recombination layer and hole transport layer thicknesses on tandem cell photocurrent generation, expe...

TT¯-deformed 2D quantum field theories

Abstract: It was noticed many years ago, in the framework of massless RG flows, that the irrelevant composite operator $$ \mathrm{T}\overline{\mathrm{T}} $$ , built with the components of the energy-momentum tensor, enjoys very special properties in 2D quantum field theories, and can be regarded as a peculiar kind of integrable perturbation. Novel interesting features of this operator have recently emerged from the study of effective string theory models. In this paper we study further properties of this distinguished perturbation. We discuss how it affects the energy levels and one-point functions of a general 2D QFT in finite volume through a surprising relation with a simple hydrodynamic equation. In the case of the perturbation of CFTs, adapting a result by Luscher and Weisz we give a compact expression for the partition function on a finite-length cylinder and make a connection with the exact g-function method. We argue that, at the classical level, the deformation naturally maps the action of N massless free bosons into the Nambu-Goto action in static gauge, in N + 2 target space dimensions, and we briefly discuss a possible interpretation of this result in the context of effective string models.

Demixed principal component analysis of neural population data

Abstract: Neurons in higher cortical areas, such as the prefrontal cortex, are often tuned to a variety of sensory and motor variables, and are therefore said to display mixed selectivity. This complexity of single neuron responses can obscure what information these areas represent and how it is represented. Here we demonstrate the advantages of a new dimensionality reduction technique, demixed principal component analysis (dPCA), that decomposes population activity into a few components. In addition to systematically capturing the majority of the variance of the data, dPCA also exposes the dependence of the neural representation on task parameters such as stimuli, decisions, or rewards. To illustrate our method we reanalyze population data from four datasets comprising different species, different cortical areas and different experimental tasks. In each case, dPCA provides a concise way of visualizing the data that summarizes the task-dependent features of the population response in a single figure.

Quench dynamics and relaxation in isolated integrable quantum spin chains

Abstract: We review the dynamics after quantum quenches in integrable quantum spin chains. We give a pedagogical introduction to relaxation in isolated quantum systems, and discuss the description of the steady state by (gen- eralized) Gibbs ensembles. When then turn to general features in the time evolution of local observables after the quench, using a simple model of free fermions as an example. In the second part we present an overview of recent progress in describing quench dynamics in two key paradigms for quantum integrable models, the transverse field Ising chain and the anisotropic spin-1/2 Heisenberg chain.

Biobased Amines: From Synthesis to Polymers; Present and Future.

Abstract: Amines are key intermediates in the chemical industry due to their nucleophilic characteristic which confers a high reactivity to them. Thus, they are key monomers for the synthesis of polyamides, polyureas, polyepoxydes, which are all of growing interest in automotive, aerospace, building, or health applications. Despite a growing interest for biobased monomers and polymers, and particularly polyamides, it should be noticed that very few natural amines are available. Actually, there is only chitosan and poly(lysine). In this review we present both fundamental and applied research on the synthesis of biobased primary and secondary amines with current available biobased resources. Their use is described as a building block for material chemistry. Hence, we first recall some background on the synthesis of amines, including the reactivity of amines. Second we focus on the synthesis of biobased amines from all sorts of biomass, from carbohydrate, terpenes, or oleochemical sources. Third, because they need opt...

Experimental Treatment with Favipiravir for Ebola Virus Disease (the JIKI Trial): A Historically Controlled, Single-Arm Proof-of-Concept Trial in Guinea.

Abstract: BACKGROUND:Ebola virus disease (EVD) is a highly lethal condition for which no specific treatment has proven efficacy. In September 2014, while the Ebola outbreak was at its peak, the World Health Organization released a short list of drugs suitable for EVD research. Favipiravir, an antiviral developed for the treatment of severe influenza, was one of these. In late 2014, the conditions for starting a randomized Ebola trial were not fulfilled for two reasons. One was the perception that, given the high number of patients presenting simultaneously and the very high mortality rate of the disease, it was ethically unacceptable to allocate patients from within the same family or village to receive or not receive an experimental drug, using a randomization process impossible to understand by very sick patients. The other was that, in the context of rumors and distrust of Ebola treatment centers, using a randomized design at the outset might lead even more patients to refuse to seek care. Therefore, we chose to conduct a multicenter non-randomized trial, in which all patients would receive favipiravir along with standardized care. The objectives of the trial were to test the feasibility and acceptability of an emergency trial in the context of a large Ebola outbreak, and to collect data on the safety and effectiveness of favipiravir in reducing mortality and viral load in patients with EVD. The trial was not aimed at directly informing future guidelines on Ebola treatment but at quickly gathering standardized preliminary data to optimize the design of future studies.METHODS AND FINDINGS:Inclusion criteria were positive Ebola virus reverse transcription PCR (RT-PCR) test, age ≥ 1 y, weight ≥ 10 kg, ability to take oral drugs, and informed consent. All participants received oral favipiravir (day 0: 6,000 mg; day 1 to day 9: 2,400 mg/d). Semi-quantitative Ebola virus RT-PCR (results expressed in "cycle threshold" [Ct]) and biochemistry tests were performed at day 0, day 2, day 4, end of symptoms, day 14, and day 30. Frozen samples were shipped to a reference biosafety level 4 laboratory for RNA viral load measurement using a quantitative reference technique (genome copies/milliliter). Outcomes were mortality, viral load evolution, and adverse events. The analysis was stratified by age and Ct value. A "target value" of mortality was defined a priori for each stratum, to guide the interpretation of interim and final analysis. Between 17 December 2014 and 8 April 2015, 126 patients were included, of whom 111 were analyzed (adults and adolescents, ≥13 y, n = 99; young children, ≤6 y, n = 12). Here we present the results obtained in the 99 adults and adolescents. Of these, 55 had a baseline Ct value ≥ 20 (Group A Ct ≥ 20), and 44 had a baseline Ct value < 20 (Group A Ct < 20). Ct values and RNA viral loads were well correlated, with Ct = 20 corresponding to RNA viral load = 7.7 log10 genome copies/ml. Mortality was 20% (95% CI 11.6%-32.4%) in Group A Ct ≥ 20 and 91% (95% CI 78.8%-91.1%) in Group A Ct < 20. Both mortality 95% CIs included the predefined target value (30% and 85%, respectively). Baseline serum creatinine was ≥110 μmol/l in 48% of patients in Group A Ct ≥ 20 (≥300 μmol/l in 14%) and in 90% of patients in Group A Ct < 20 (≥300 μmol/l in 44%). In Group A Ct ≥ 20, 17% of patients with baseline creatinine ≥110 μmol/l died, versus 97% in Group A Ct < 20. In patients who survived, the mean decrease in viral load was 0.33 log10 copies/ml per day of follow-up. RNA viral load values and mortality were not significantly different between adults starting favipiravir within <72 h of symptoms compared to others. Favipiravir was well tolerated.CONCLUSIONS:In the context of an outbreak at its peak, with crowded care centers, randomizing patients to receive either standard care or standard care plus an experimental drug was not felt to be appropriate. We did a non-randomized trial. This trial reaches nuanced conclusions. On the one hand, we do not conclude on the efficacy of the drug, and our conclusions on tolerance, although encouraging, are not as firm as they could have been if we had used randomization. On the other hand, we learned about how to quickly set up and run an Ebola trial, in close relationship with the community and non-governmental organizations; we integrated research into care so that it improved care; and we generated knowledge on EVD that is useful to further research. Our data illustrate the frequency of renal dysfunction and the powerful prognostic value of low Ct values. They suggest that drug trials in EVD should systematically stratify analyses by baseline Ct value, as a surrogate of viral load. They also suggest that favipiravir monotherapy merits further study in patients with medium to high viremia, but not in those with very high viremia.TRIAL REGISTRATION:ClinicalTrials.gov NCT02329054.

A dynamic model and an algorithm for short-term supply chain scheduling in the smart factory industry 4.0

Abstract: Smart factories Industry 4.0 on the basis of collaborative cyber-physical systems represents a future form of industrial networks. Supply chains in such networks have dynamic structures which evolve over time. In these settings, short-term supply chain scheduling in smart factories Industry 4.0 is challenged by temporal machine structures, different processing speed at parallel machines and dynamic job arrivals. In this study, for the first time, a dynamic model and algorithm for short-term supply chain scheduling in smart factories Industry 4.0 is presented. The peculiarity of the considered problem is the simultaneous consideration of both machine structure selection and job assignments. The scheduling approach is based on a dynamic non-stationary interpretation of the execution of the jobs and a temporal decomposition of the scheduling problem. The algorithmic realisation is based on a modified form of the continuous maximum principle blended with mathematical optimisation. A detailed theoretical analy...

Efficient codes and balanced networks

Abstract: Recent years have seen a growing interest in inhibitory interneurons and their circuits. A striking property of cortical inhibition is how tightly it balances excitation. Inhibitory currents not only match excitatory currents on average, but track them on a millisecond time scale, whether they are caused by external stimuli or spontaneous fluctuations. We review, together with experimental evidence, recent theoretical approaches that investigate the advantages of such tight balance for coding and computation. These studies suggest a possible revision of the dominant view that neurons represent information with firing rates corrupted by Poisson noise. Instead, tight excitatory/inhibitory balance may be a signature of a highly cooperative code, orders of magnitude more precise than a Poisson rate code. Moreover, tight balance may provide a template that allows cortical neurons to construct high-dimensional population codes and learn complex functions of their inputs.

Assessing the Ability of LSTMs to Learn Syntax-Sensitive Dependencies

Abstract: The success of long short-term memory (LSTM) neural networks in language processing is typically attributed to their ability to capture long-distance statistical regularities. Linguistic regularities are often sensitive to syntactic structure; can such dependencies be captured by LSTMs, which do not have explicit structural representations? We begin addressing this question using number agreement in English subject-verb dependencies. We probe the architecture's grammatical competence both using training objectives with an explicit grammatical target (number prediction, grammaticality judgments) and using language models. In the strongly supervised settings, the LSTM achieved very high overall accuracy (less than 1% errors), but errors increased when sequential and structural information conflicted. The frequency of such errors rose sharply in the language-modeling setting. We conclude that LSTMs can capture a non-trivial amount of grammatical structure given targeted supervision, but stronger architectures may be required to further reduce errors; furthermore, the language modeling signal is insufficient for capturing syntax-sensitive dependencies, and should be supplemented with more direct supervision if such dependencies need to be captured.

Consequences of twenty-first-century policy for multi-millennial climate and sea-level change

Abstract: Most of the policy debate surrounding the actions needed to mitigate and adapt to anthropogenic climate change has been framed by observations of the past 150 years as well as climate and sea-level projections for the twenty-first century. The focus on this 250-year window, however, obscures some of the most profound problems associated with climate change. Here, we argue that the twentieth and twenty-first centuries, a period during which the overwhelming majority of human-caused carbon emissions are likely to occur, need to be placed into a long-term context that includes the past 20 millennia, when the last Ice Age ended and human civilization developed, and the next ten millennia, over which time the projected impacts of anthropogenic climate change will grow and persist. This long-term perspective illustrates that policy decisions made in the next few years to decades will have profound impacts on global climate, ecosystems and human societies — not just for this century, but for the next ten millennia and beyond.

Brownian Carnot engine

Abstract: The Carnot cycle imposes a fundamental upper limit to the efficiency of a macroscopic motor operating between two thermal baths. However, this bound needs to be reinterpreted at microscopic scales, where molecular bio-motors and some artificial micro-engines operate. As described by stochastic thermodynamics, energy transfers in microscopic systems are random and thermal fluctuations induce transient decreases of entropy, allowing for possible violations of the Carnot limit. Here we report an experimental realization of a Carnot engine with a single optically trapped Brownian particle as the working substance. We present an exhaustive study of the energetics of the engine and analyse the fluctuations of the finite-time efficiency, showing that the Carnot bound can be surpassed for a small number of non-equilibrium cycles. As its macroscopic counterpart, the energetics of our Carnot device exhibits basic properties that one would expect to observe in any microscopic energy transducer operating with baths at different temperatures. Our results characterize the sources of irreversibility in the engine and the statistical properties of the efficiency-an insight that could inspire new strategies in the design of efficient nano-motors.

Efficient Near-Infrared-Transparent Perovskite Solar Cells Enabling Direct Comparison of 4-Terminal and Monolithic Perovskite/Silicon Tandem Cells

Abstract: Combining market-proven silicon solar cell technology with an efficient wide band gap top cell into a tandem device is an attractive approach to reduce the cost of photovoltaic systems. For this, perovskite solar cells are promising high-efficiency top cell candidates, but their typical device size (<0.2 cm2), is still far from standard industrial sizes. We present a 1 cm2 near-infrared transparent perovskite solar cell with 14.5% steady-state efficiency, as compared to 16.4% on 0.25 cm2. By mechanically stacking these cells with silicon heterojunction cells, we experimentally demonstrate a 4-terminal tandem measurement with a steady-state efficiency of 25.2%, with a 0.25 cm2 top cell. The developed top cell processing methods enable the fabrication of a 20.5% efficient and 1.43 cm2 large monolithic perovskite/silicon heterojunction tandem solar cell, featuring a rear-side textured bottom cell to increase its near-infrared spectral response. Finally, we compare both tandem configurations to identify effic...

Storm track processes and the opposing influences of climate change

Abstract: Extratropical storms contribute to precipitation, wind and temperature extremes. A synthesis of the influences of a changing climate on storm tracks reveals competing effects on meridional temperature gradients, which make projections difficult.

Learning Text Similarity with Siamese Recurrent Networks

Abstract: This paper presents a deep architecture for learning a similarity metric on variablelength character sequences. The model combines a stack of character-level bidirectional LSTM’s with a Siamese architecture. It learns to project variablelength strings into a fixed-dimensional embedding space by using only information about the similarity between pairs of strings. This model is applied to the task of job title normalization based on a manually annotated taxonomy. A small data set is incrementally expanded and augmented with new sources of variance. The model learns a representation that is selective to differences in the input that reflect semantic differences (e.g., “Java developer” vs. “HR manager”) but also invariant to nonsemantic string differences (e.g., “Java developer” vs. “Java programmer”).

Structure of the E6/E6AP/p53 complex required for HPV-mediated degradation of p53

Abstract: The p53 pro-apoptotic tumour suppressor is mutated or functionally altered in most cancers. In epithelial tumours induced by 'high-risk' mucosal human papilloma viruses, including human cervical carcinoma and a growing number of head-and-neck cancers, p53 is degraded by the viral oncoprotein E6 (ref. 2). In this process, E6 binds to a short leucine (L)-rich LxxLL consensus sequence within the cellular ubiquitin ligase E6AP. Subsequently, the E6/E6AP heterodimer recruits and degrades p53 (ref. 4). Neither E6 nor E6AP are separately able to recruit p53 (refs 3, 5), and the precise mode of assembly of E6, E6AP and p53 is unknown. Here we solve the crystal structure of a ternary complex comprising full-length human papilloma virus type 16 (HPV-16) E6, the LxxLL motif of E6AP and the core domain of p53. The LxxLL motif of E6AP renders the conformation of E6 competent for interaction with p53 by structuring a p53-binding cleft on E6. Mutagenesis of critical positions at the E6-p53 interface disrupts p53 degradation. The E6-binding site of p53 is distal from previously described DNA- and protein-binding surfaces of the core domain. This suggests that, in principle, E6 may avoid competition with cellular factors by targeting both free and bound p53 molecules. The E6/E6AP/p53 complex represents a prototype of viral hijacking of both the ubiquitin-mediated protein degradation pathway and the p53 tumour suppressor pathway. The present structure provides a framework for the design of inhibitory therapeutic strategies against oncogenesis mediated by human papilloma virus.

Robustness of classifiers: from adversarial to random noise

Abstract: Several recent works have shown that state-of-the-art classifiers are vulnerable to worst-case (i.e., adversarial) perturbations of the datapoints. On the other hand, it has been empirically observed that these same classifiers are relatively robust to random noise. In this paper, we propose to study a semi-random noise regime that generalizes both the random and worst-case noise regimes. We propose the first quantitative analysis of the robustness of nonlinear classifiers in this general noise regime. We establish precise theoretical bounds on the robustness of classifiers in this general regime, which depend on the curvature of the classifier's decision boundary. Our bounds confirm and quantify the empirical observations that classifiers satisfying curvature constraints are robust to random noise. Moreover, we quantify the robustness of classifiers in terms of the subspace dimension in the semi-random noise regime, and show that our bounds remarkably interpolate between the worst-case and random noise regimes. We perform experiments and show that the derived bounds provide very accurate estimates when applied to various state-of-the-art deep neural networks and datasets. This result suggests bounds on the curvature of the classifiers' decision boundaries that we support experimentally, and more generally offers important insights onto the geometry of high dimensional classification problems.