Search

With the growing success of Large Language models (LLMs) in information-seeking scenarios, search engines are now adopting generative approaches to provide answers along with in-line citations as attribution. While existing work focuses mainly on attributed question answering, in this paper, we target information-seeking scenarios which are often more challenging due to the open-ended nature of the queries and the size of the label space in terms of the diversity of candidate-attributed answers per query. We propose a reproducible framework to evaluate and benchmark attributed information seeking, using any backbone LLM, and different architectural designs: (1) Generate (2) Retrieve then Generate, and (3) Generate then Retrieve. Experiments using HAGRID, an attributed information-seeking dataset, show the impact of different scenarios on both the correctness and attributability of answers.

Knowledge of the charge-carrier recombination lifetime, tau, is crucial for the various applications of photovoltaic perovskites. We studied the novel inorganic perovskite, CsPbBr3 and we observe recombination dynamics beyond 1 ms below 200 K and tau approaching 100 us at room temperature. Time-resolved microwave-detected photoconductivity decay (TRMCD), used in combination with injection dependence, evidence that tau is dominated by impurity charge trapping. The observed injection dependence is well corroborated by modeling of the trap mechanism. The ultra-long decay time is also consistent with photoconductivity measurements with a continuous-wave excitation at powers corresponding to around one Sun irradiation. While in principle charge-carrier trapping may limit the photovoltaic efficiency in single-cell photovoltaic devices, it could also lead to enhanced efficiency in tandem cells as well as for alternative applications including photodetection and quantum information storage., Comment: 16 pages, 5 figures, submitted to ACS Photonics

The number of cores on graphical computing units (GPUs) is reaching thousands nowadays, whereas the clock speed of processors stagnates. Unfortunately, constraint programming solvers do not take advantage yet of GPU parallelism. One reason is that constraint solvers were primarily designed within the mental frame of sequential computation. To solve this issue, we take a step back and contribute to a simple, intrinsically parallel, lock-free and formally correct programming language based on concurrent constraint programming. We then re-examine parallel constraint solving on GPUs within this formalism, and develop Turbo, a simple constraint solver entirely programmed on GPUs. Turbo validates the correctness of our approach and compares positively to a parallel CPU-based solver., Comment: AAAI-2022 paper with erratum in Section "Turbo: GPU Constraint Solver"

In this work, our aim is to provide a structured answer in natural language to a complex information need. Particularly, we envision using generative models from the perspective of data-to-text generation. We propose the use of a content selection and planning pipeline which aims at structuring the answer by generating intermediate plans. The experimental evaluation is performed using the TREC Complex Answer Retrieval (CAR) dataset. We evaluate both the generated answer and its corresponding structure and show the effectiveness of planning-based models in comparison to a text-to-text model., Comment: 8 pages, 1 figure, ECIR 2022 short paper

In this work, we present the Deep Newton Reconstruction Network (DNR-Net), a hybrid data-driven reconstruction technique for emission tomography inspired by Newton's method, a well-known iterative optimization algorithm. The DNR-Net employs prior information about the tomographic problem provided by the projection operator while utilizing deep learning approaches to a) imitate Newton's method by approximating the Newton descent direction and b) provide data-driven regularisation. We demonstrate that DNR-Net is capable of providing high-quality image reconstructions using data from SPECT phantom simulations by applying it to reconstruct images from noisy sinograms, each one containing 24 projections. The Structural Similarity Index (SSIM) and the Contrast-to-Noise ratio (CNR) were used to quantify the image quality. We also compare our results to those obtained by the OSEM method. According to the quantitative results, the DNR-Net produces reconstructions comparable to the ones produced by OSEM while featuring higher contrast and less noise., Comment: 7 pages, 4 figures, Proceedings

Results from the TESS mission showed that previous studies strngly underestimated the number of slow rotators, revealing the importance of studying those asteroids. For most slowly rotating asteroids (P > 12), no spin and shape model is available because of observation selection effects. This hampers determination of their thermal parameters and accurate sizes. We continue our campaign in minimising selection effects among main belt asteroids. Our targets are slow rotators with low light-curve amplitudes. The goal is to provide their scaled spin and shape models together with thermal inertia, albedo, and surface roughness to complete the statistics. Rich multi-apparition datasets of dense light curves are supplemented with data from Kepler and TESS. In addition to data in the visible range, we also use thermal data from infrared space observatories (IRAS, Akari and WISE) in a combined optimisation process using the Convex Inversion Thermophysical Model (CITPM). This novel method has so far been applied to only a few targets, and in this work we further validate the method. We present the models of 16 slow rotators. All provide good fits to both thermal and visible data. The obtained sizes are on average accurate at the 5% precision, with diameters in the range from 25 to 145 km. The rotation periods of our targets range from 11 to 59 hours, and the thermal inertia covers a wide range of values, from 2 to <400 SI units, not showing any correlation with the period. With this work we increase the sample of slow rotators with reliable spin and shape models and known thermal inertia by 40%. The thermal inertia values of our sample do not display a previously suggested increasing trend with rotation period, which might be due to their small skin depth., Comment: Accepted to Astronomy & Astrophysics. 10 pages + appendices

The development of deep neural networks and the emergence of pre-trained language models such as BERT allow to increase performance on many NLP tasks. However, these models do not meet the same popularity for tweet summarization, which can probably be explained by the lack of existing collections for training and evaluation. Our contribution in this paper is twofold : (1) we introduce a large dataset for Twitter event summarization, and (2) we propose a neural model to automatically summarize huge tweet streams. This extractive model combines in an original way pre-trained language models and vocabulary frequency-based representations to predict tweet salience. An additional advantage of the model is that it automatically adapts the size of the output summary according to the input tweet stream. We conducted experiments using two different Twitter collections, and promising results are observed in comparison with state-of-the-art baselines., Comment: UNDER SUBMISSION

With the widespread use of toxic language online, platforms are increasingly using automated systems that leverage advances in natural language processing to automatically flag and remove toxic comments. However, most automated systems -- when detecting and moderating toxic language -- do not provide feedback to their users, let alone provide an avenue of recourse for these users to make actionable changes. We present our work, RECAST, an interactive, open-sourced web tool for visualizing these models' toxic predictions, while providing alternative suggestions for flagged toxic language. Our work also provides users with a new path of recourse when using these automated moderation tools. RECAST highlights text responsible for classifying toxicity, and allows users to interactively substitute potentially toxic phrases with neutral alternatives. We examined the effect of RECAST via two large-scale user evaluations, and found that RECAST was highly effective at helping users reduce toxicity as detected through the model. Users also gained a stronger understanding of the underlying toxicity criterion used by black-box models, enabling transparency and recourse. In addition, we found that when users focus on optimizing language for these models instead of their own judgement (which is the implied incentive and goal of deploying automated models), these models cease to be effective classifiers of toxicity compared to human annotations. This opens a discussion for how toxicity detection models work and should work, and their effect on the future of online discourse., Comment: 26 pages, 5 figures, CSCW '21

Several deep neural ranking models have been proposed in the recent IR literature. While their transferability to one target domain held by a dataset has been widely addressed using traditional domain adaptation strategies, the question of their cross-domain transferability is still under-studied. We study here in what extent neural ranking models catastrophically forget old knowledge acquired from previously observed domains after acquiring new knowledge, leading to performance decrease on those domains. Our experiments show that the effectiveness of neuralIR ranking models is achieved at the cost of catastrophic forgetting and that a lifelong learning strategy using a cross-domain regularizer success-fully mitigates the problem. Using an explanatory approach built on a regression model, we also show the effect of domain characteristics on the rise of catastrophic forgetting. We believe that the obtained results can be useful for both theoretical and practical future work in neural IR.

In this work, we study the effect of beam deviation angle at the reception side and calculate the theoretical demultiplexed collected energy for up to 15 modes, investigating the influence of the ratio between incoming beam size and fundamental HG mode's waist. We show this approach greatly enhances the collection efficiency, tolerating tip-tilt error of more than 3 times compared to a Gaussian beam alone. We also show that, depending on wait size, a trade-off between collection efficiency at small angles and maximum acceptance angle can be achieved., Comment: 4 pages, 11 figures, conference

Recent seroprevalence studies have tried to estimate the real number of asymptomatic cases affected by COVID-19. It is of paramount importance to understand the impact of these infections in order to prevent a second wave. This study aims to model the interactions in the population by means of a complex network and to shed some light on the effectiveness of localised control measures in Italy in relation to the school opening in mid-September. The formulation of an epidemiological predictive model is given: the advantage of using this model lies in that it discriminates between asymptomatic and symptomatic cases of COVID-19 as the interactions with these two categories of infected individuals are captured separately, allowing for a study on the impact of asymptomatic cases. This model is then extended to a structured nonhomogeneous version by means of the Watts-Strogatz complex network, which is adopted widely to model societal interactions as it holds the small world property. Finally, a case study on the situation in Italy is given: first the homogeneous model is used to compare the official data with the data of the recent seroprevalence study from Istat; second, in view of the return to school in mid-September, a study at regional level is conducted. The results of this study highlight the importance of coordinating the deployment of appropriate control measures that take into account the role of asymptomatic infections, especially in younger individuals, and inter-regional connectivity in Italy., Comment: 38 pages, journal

This paper presents new methods for ensuring that the energy system of a neighborhood that is designed with the objective of being zero emission is actually operated in a way that allows it to reach net zero emissions in its lifetime. This paper highlights the necessity of taking into account realistic operation strategies when designing the energy system of such neighborhoods. It also suggests methods that can be used in the operation of ZENs to ensure carbon neutrality. An optimization model for designing the energy system of a Zero Emission Neighborhood (ZEN) is first presented and used to produce two designs for a campus in the South of Norway in the case where the amount of PV is limited (PVlim) and when it is not (Base). Several operation approaches are then introduced to compare their operation cost and the CO2 emissions and compensations. These approaches are perfect foresight used as a reference (Ref.),a purely economic model predictive control (E-MPC), an MPC with penalization if deviating from emission targets (EmE-MPC)and a receding horizon MPC where we have a net zero emission constraint over the year (RH-MPC). The resulting energy systems are, in the Base case, PV, heat pumps, a gas boiler and heat storage and, in the PVlim case, a smaller amount of PV, a CHP plant, and heat storage. In the Base case all operation strategies manage to reach net zero emissions, largely due to the passive compensations obtained from the PV. RH-MPC offers the lowest cost. In the PVlim case, the passive effect of the PV is not sufficient to reach net zero emissions and an operation approach specifically taking into account the emissions is necessary. EmE-MPC achieves the lowest emissions but it comes at a much higher cost. We conclude that the best overall strategy is RH-MPC which maintains both the cost and the emission-compensation balance close to the reference case with perfect foresight.

This paper investigates the use of clustering in the context of designing the energy system of Zero Emission Neighborhoods (ZEN). ZENs are neighborhoods who aim to have net zero emissions during their lifetime. While previous work has used and studied clustering for designing the energy system of neighborhoods, no article dealt with neighborhoods such as ZEN, which have high requirements for the solar irradiance time series, include a CO2 factor time series and have a zero emission balance limiting the possibilities. To this end several methods are used and their results compared. The results are on the one hand the performances of the clustering itself and on the other hand, the performances of each method in the optimization model where the data is used. Various aspects related to the clustering methods are tested. The different aspects studied are: the goal (clustering to obtain days or hours), the algorithm (k-means or k-medoids), the normalization method (based on the standard deviation or range of values) and the use of heuristic. The results highlight that k-means offers better results than k-medoids and that k-means was systematically underestimating the objective value while k-medoids was constantly overestimating it. When the choice between clustering days and hours is possible, it appears that clustering days offers the best precision and solving time. The choice depends on the formulation used for the optimization model and the need to model seasonal storage. The choice of the normalization method has the least impact, but the range of values method show some advantages in terms of solving time. When a good representation of the solar irradiance time series is needed, a higher number of days or using hours is necessary. The choice depends on what solving time is acceptable., Comment: 12 pages, 19 figures, 7 tables, 1 Appendix

As toxic language becomes nearly pervasive online, there has been increasing interest in leveraging the advancements in natural language processing (NLP), from very large transformer models to automatically detecting and removing toxic comments. Despite the fairness concerns, lack of adversarial robustness, and limited prediction explainability for deep learning systems, there is currently little work for auditing these systems and understanding how they work for both developers and users. We present our ongoing work, RECAST, an interactive tool for examining toxicity detection models by visualizing explanations for predictions and providing alternative wordings for detected toxic speech., Comment: 8 Pages, 3 figures, The eighth International Workshop of Chinese CHI Proceedings

This paper investigates the relationship between grid tariffs and investment in Zero Emission Neighborhoods (ZEN) energy system, and how the grid exchanges are affected. Different grid tariffs (energy based, time of use (ToU), subscribed capacity and dynamic) are implemented in an optimization model that minimizes the cost of investing and operating a ZEN during its lifetime. The analysis is conducted in two cases: non-constrained exports and exports limited to 100kWh/h. The results suggest that in the case with no limit on export, the grid tariff has little influence, but ToU is economically advantageous for both the ZEN and the DSO. When exports are limited, the subscribed capacity scheme allows to maintain DSO revenue, while the others cut them by half. This tariff also offers the lowest maximum peak and a good duration curve. The dynamic tariff creates new potentially problematic peak imports despite its benefits in other peak hours., Comment: Presented at IEEE Powertech 2019 in Milano, 6 pages

This project examined perceptions of the vegan lifestyle using surveys and social media to explore barriers to choosing veganism. A survey of 510 individuals indicated that non-vegans did not believe veganism was as healthy or difficult as vegans. In a second analysis, Instagram posts using #vegan suggest content is aimed primarily at the female vegan community. Finally, sentiment analysis of roughly 5 million Twitter posts mentioning 'vegan' found veganism to be portrayed in a more positive light compared to other topics. Results suggest non-vegans' lack of interest in veganism is driven by non-belief in the health benefits of the diet.

Zero Emission Neighborhoods (ZEN) is a concept studied in Norway to reduce the CO_2 emission of neighborhoods. One question coming along this concept is how to design the energy system of such neighborhoods to fit the ZEN definition. From this definition we extract the CO_2 balance, requiring an annual net zero emission of CO_2 in the lifetime of the neighborhood. This paper proposes a MILP model for obtaining cost optimal design of ZEN's energy system and demonstrates it on a case study. Different technologies are included as investment options and, notably PV as an on-site electricity production mean. Wind turbines are not included in this study because inappropriate in the context of most cities. The results for the case study highlight the importance of PV investment in reaching the ZEN requirements. For example, around 850 kW of solar is needed for our test cases of 10 000 m^2 of floor area, for an annual energy demand of around 700 MWh of electricity and 620 MWh of heat. The investments in other technologies are small in comparison.

Space Division Multiplexing (SMD) is a very attractive technique for addressing the ever-growing demands in transmission capacity by enabling the use of a new parameter \textemdash\ space \textemdash\ to increase the number of channels in multi-mode fibers. One key component to build a spatially multiplexed-based optical network is a spatial multiplexer and demultiplexer combining signals from multiple single-mode fibers into as many channels in a multi-mode fiber. In this article, we report the fabrication and characterization of a pair of 45-mode spatial multiplexer and demultiplexer saturating all the modes of a standard 50~$\mu$m core graded-index (OM2) multi-mode fiber. The multiplexers are based on Multi-Plane Light Conversion (MPLC), a technique that enables the control of the transverse shape of the light by multiple reflections on specifically designed phase plates. We show that by using a separable variable basis of modes, such as Hermite-Gaussian (HG) modes, we are able to drastically reduce the number of reflections hence reducing the insertion losses and modal crosstalks. The multiplexers typically show an average 4~dB insertion loss and -28~dB cross-talk across the C band. Finally, we emphasize the use of this higher-order modes multiplexer to explore the propagation properties inside multi-mode fibers and more specifically the mode group crosstalks as well as the impact of fiber bending., Comment: Follow-up of OFC PDP publication. Upgraded version

The TREC Real-Time Summarization (RTS) track provides a framework for evaluating systems monitoring the Twitter stream and pushing tweets to users according to given profiles. It includes metrics, files, settings and hypothesis provided by the organizers. In this work, we perform a thorough analysis of each component of the framework used in 2016 and 2017 and found some limitations for the Scenario A of this track. Our main findings point out the weakness of the metrics and give clear recommendations to fairly reuse the collection.

Euclid is a space-based optical/near-infrared survey mission of the European Space Agency (ESA) to investigate the nature of dark energy, dark matter and gravity by observing the geometry of the Universe and on the formation of structures over cosmological timescales. Euclid will use two probes of the signature of dark matter and energy: Weak gravitational Lensing, which requires the measurement of the shape and photometric redshifts of distant galaxies, and Galaxy Clustering, based on the measurement of the 3-dimensional distribution of galaxies through their spectroscopic redshifts. The mission is scheduled for launch in 2020 and is designed for 6 years of nominal survey operations. The Euclid Spacecraft is composed of a Service Module and a Payload Module. The Service Module comprises all the conventional spacecraft subsystems, the instruments warm electronics units, the sun shield and the solar arrays. In particular the Service Module provides the extremely challenging pointing accuracy required by the scientific objectives. The Payload Module consists of a 1.2 m three-mirror Korsch type telescope and of two instruments, the visible imager and the near-infrared spectro-photometer, both covering a large common field-of-view enabling to survey more than 35% of the entire sky. All sensor data are downlinked using K-band transmission and processed by a dedicated ground segment for science data processing. The Euclid data and catalogues will be made available to the public at the ESA Science Data Centre., Comment: 23 pages, 19 figures, Presented at the SPIE Astronomical Telescopes and Instrumentation conference in Edinburgh, Scotland, United Kingdom, 6 June 1 July 2016

Precise control of atom-light interactions is vital to many quantum information protocols. In particular, atomic systems can be used to slow and store light to form a quantum memory. Optical storage can be achieved via stopped light, where no optical energy remains in the atoms, or as stationary light, where some optical energy remains resent during storage. In this work, we demonstrate a form of self-stabilising stationary light. From any initial state, our atom-light system evolves to a stable configuration that is devoid of coherent emission from the atoms, yet may contain bright optical excitation. This phenomenon is verified experimentally in a cloud of cold Rb87 atoms. The spinwave in our atomic cloud is imaged from the side allowing direct comparison with theoretical predictions.

Optical resonance is central to a wide range of optical devices and techniques. In an optical cavity, the round-trip length and mirror reflectivity can be chosen to optimize the circulating optical power, linewidth, and free-spectral range (FSR) for a given application. In this paper we show how an atomic spinwave system, with no physical mirrors, can behave in a manner that is analogous to an optical cavity. We demonstrate this similarity by characterising the build-up and decay of the resonance in the time domain, and measuring the effective optical linewidth and FSR in the frequency domain. Our spinwave is generated in a 20 cm long Rb gas cell, yet it facilitates an effective FSR of 83 kHz, which would require a round-trip path of 3.6 km in a free-space optical cavity. Furthermore, the spinwave coupling is controllable enabling dynamic tuning of the effective cavity parameters., Comment: 13 pages, 4 figures

Exploiting 2D materials for spintronic applications can potentially realize next-generation devices featuring low-power consumption and quantum operation capability. The magnetic exchange field (MEF) induced by an adjacent magnetic insulator enables efficient control of local spin generation and spin modulation in 2D devices without compromising the delicate material structures. Using graphene as a prototypical 2D system, we demonstrate that its coupling to the model magnetic insulator (EuS) produces a substantial MEF (> 14 T) with potential to reach hundreds of Tesla, which leads to orders-of-magnitude enhancement in the spin signal originated from Zeeman spin-Hall effect. Furthermore, the new ferromagnetic ground state of Dirac electrons resulting from the strong MEF may give rise to quantized spin-polarized edge transport. The MEF effect shown in our graphene/EuS devices therefore provides a key functionality for future spin logic and memory devices based on emerging 2D materials in classical and quantum information processing.

Photon-subtracted and photon-added Gaussian states are amongst the simplest non-Gaussian states that are experimentally available. It is generally believed that they are some of the best candidates to enhance sensitivity in parameter extraction. We derive here the quantum Cram\'er-Rao bound for such states and find that for large photon numbers photon-subtraction or -addition only leads to a small correction of the quantum Fisher information (QFI). On the other hand a divergence of the QFI appears for very small squeezing in the limit of vanishing photon number in the case of photon subtraction, implying an arbitrarily precise measurement with almost no light. However, at least for the standard and experimentally established preparation scheme, the decreasing success probability of the preparation in that limit exactly cancels the divergence, leading to finite sensitivity per square root of Hertz, when the duration of the preparation is taken into account., Comment: 19 pages, 3 figures

We show that a set of optical memories can act as a configurable linear optical network operating on frequency-multiplexed optical states. Our protocol is applicable to any quantum memories that employ off-resonant Raman transitions to store optical information in atomic spins. In addition to the configurability, the protocol also offers favourable scaling with an increasing number of modes where N memories can be configured to implement an arbitrary N-mode unitary operations during storage and readout. We demonstrate the versatility of this protocol by showing an example where cascaded memories are used to implement a conditional CZ gate., Comment: 5 pages, 2 figures

Computational creativity is an emerging branch of artificial intelligence that places computers in the center of the creative process. Broadly, creativity involves a generative step to produce many ideas and a selective step to determine the ones that are the best. Many previous attempts at computational creativity, however, have not been able to achieve a valid selective step. This work shows how bringing data sources from the creative domain and from hedonic psychophysics together with big data analytics techniques can overcome this shortcoming to yield a system that can produce novel and high-quality creative artifacts. Our data-driven approach is demonstrated through a computational creativity system for culinary recipes and menus we developed and deployed, which can operate either autonomously or semi-autonomously with human interaction. We also comment on the volume, velocity, variety, and veracity of data in computational creativity.

A method for time differentiation based on a Babinet-Soleil-Bravais compensator is introduced. The complex transfer function of the device is measured using polarization spectral interferometry. Time differentiation of both the pulse field and pulse envelope are demonstrated over a spectral width of about 100 THz with a measured overlap with the objective mode greater than 99.8%. This pulse shaping technique is shown to be perfectly suited to time metrology at the quantum limit.

We calculate the quantum Cram\'er--Rao bound for the sensitivity with which one or several parameters, encoded in a general single-mode Gaussian state, can be estimated. This includes in particular the interesting case of mixed Gaussian states. We apply the formula to the problems of estimating phase, purity, loss, amplitude, and squeezing. In the case of the simultaneous measurement of several parameters, we provide the full quantum Fisher information matrix. Our results unify previously known partial results, and constitute a complete solution to the problem of knowing the best possible sensitivity of measurements based on a single-mode Gaussian state., Comment: This paper has been withdrawn because of a double submission. See: http://arxiv.org/abs/1307.5318

Typical cohorts in brain imaging studies are not large enough for systematic testing of all the information contained in the images. To build testable working hypotheses, investigators thus rely on analysis of previous work, sometimes formalized in a so-called meta-analysis. In brain imaging, this approach underlies the specification of regions of interest (ROIs) that are usually selected on the basis of the coordinates of previously detected effects. In this paper, we propose to use a database of images, rather than coordinates, and frame the problem as transfer learning: learning a discriminant model on a reference task to apply it to a different but related new task. To facilitate statistical analysis of small cohorts, we use a sparse discriminant model that selects predictive voxels on the reference task and thus provides a principled procedure to define ROIs. The benefits of our approach are twofold. First it uses the reference database for prediction, i.e. to provide potential biomarkers in a clinical setting. Second it increases statistical power on the new task. We demonstrate on a set of 18 pairs of functional MRI experimental conditions that our approach gives good prediction. In addition, on a specific transfer situation involving different scanners at different locations, we show that voxel selection based on transfer learning leads to higher detection power on small cohorts., Comment: MICCAI, Nice : France (2012)

We propose a direct and real-time ranging scheme using an optical frequency combs, able to compensate optically for index of refraction variations due to atmospheric parameters. This scheme could be useful for applications requiring stringent precision over a long distance in air, a situation where dispersion becomes the main limitation. The key ingredient is the use of a mode-locked laser as a precise source for multi-wavelength interferometry in a homodyne detection scheme. By shaping temporally the local oscillator, one can directly access the desired parameter (distance) while being insensitive to fluctuations induced by parameters of the environment such as pressure, temperature, humidity and CO$_2$ content.

Three-level atomic gradient echo memory (lambda-GEM) is a proposed candidate for efficient quantum storage and for linear optical quantum computation with time-bin multiplexing. In this paper we investigate the spatial multimode properties of a lambda-GEM system. Using a high-speed triggered CCD, we demonstrate the storage of complex spatial modes and images. We also present an in-principle demonstration of spatial multiplexing by showing selective recall of spatial elements of a stored spin wave. Using our measurements, we consider the effect of diffusion within the atomic vapour and investigate its role in spatial decoherence. Our measurements allow us to quantify the spatial distortion due to both diffusion and inhomogeneous control field scattering and compare these to theoretical models., Comment: 11 pages, 9 figures