Subsidence, the lowering of Earth's land surface, is a potentially destructive hazard that can be caused by a wide range of natural or anthropogenic triggers but mainly results from solid or fluid mobilization underground. Subsidence due to groundwater depletion ([ 1 ][1]) is a slow and gradual process that develops on large time scales (months to years), producing progressive loss of land elevation (centimeters to decimeters per year) typically over very large areas (tens to thousands of square kilometers) and variably affects urban and agricultural areas worldwide. Subsidence permanently reduces aquifer-system storage capacity, causes earth fissures, damages buildings and civil infrastructure, and increases flood susceptibility and risk. During the next decades, global population and economic growth will continue to increase groundwater demand and accompanying groundwater depletion ([ 2 ][2]) and, when exacerbated by droughts ([ 3 ][3]), will probably increase land subsidence occurrence and related damages or impacts. To raise awareness and inform decision-making, we evaluate potential global subsidence due to groundwater depletion, a key first step toward formulating effective land-subsidence policies that are lacking in most countries worldwide. A large-scale systematic literature review reveals that during the past century, land subsidence due to groundwater depletion occurred at 200 locations in 34 countries [see supplementary materials (SM)]. However, subsidence extent is only known for one-third of these records, information on the impacts is scarce, and mitigation measures were implemented only in a few locations. In China, widespread subsidence affects cities developed in the main sedimentary basins. In Indonesia, coastal subsidence in Jakarta is so severe that government authorities are planning to move the capital to the island of Borneo. In Japan, subsidence affected several cities during the 20th century, including more than 4 m of subsidence in Tokyo, before groundwater management practices mitigated further subsidence. Iran currently hosts some of the fastest-sinking cities in the world (25 cm year--1) because of unregulated groundwater pumping. In Europe, the greatest impact of subsidence occurs in the Netherlands, where subsidence is primarily responsible for placing 25% of the country below the mean sea level and increasing the flooding risk. Subsidence in the Po River Plain in Italy started during the second half of the 20th century and currently threatens 30% of the Italian population, contributing to recurrent coastal flooding during extreme high tides in Venice. In North America, intense groundwater depletion triggers subsidence from California's Central Valley, with as much as 9 m of subsidence in the past century, to the Atlantic and Gulf of Mexico coastal plains in the United States, where subsidence is increasing flooding risk. In México, subsidence rates are among the highest worldwide (as much as 30 cm year-1), affecting small structurally controlled intermontane basins where the main urban centers developed, causing an important but unaccounted economic impact. Spatial analysis of subsidence locations identified in our global database (see SM) reveals that subsidence has preferentially occurred in very flat areas where unconsolidated sediments accumulated in alluvial basins or coastal plains, and where urban or agricultural areas developed in temperate or arid climates characterized by prolonged dry periods. Land subsidence has generally occurred in water-stressed basins, where the combination of groundwater withdrawal and natural groundwater discharge outpaced groundwater recharge, resulting in groundwater storage losses, groundwater depletion, and compaction of susceptible aquifer systems. In the affected basins, land subsidence mainly occurred in highly populated areas, with half of documented occurrences in areas susceptible to flooding. In coastal zones, the combined effects of absolute sea-level rise and land subsidence contribute to relative sea-level rise ([ 4 ][4]). The contribution from land subsidence may exceed the contribution from absolute sea-level rise by a factor of 10 or more and could be especially critical for 21% of the geographic locations identified in our database, where land elevation is less than 1 m above the mean sea level. On the basis of the spatial analysis findings, a global model is proposed to combine the main variables influencing subsidence to identify environmental settings favoring land subsidence and the anthropogenic factors leading to groundwater depletion (see SM). Statistical analyses of lithology, land-surface slope, land cover, and Koppen-Geiger climate classes are used to predict global subsidence susceptibility at a spatial resolution of 1 km2. The probability of groundwater depletion is estimated by identifying urban and irrigated areas suffering water stress and where groundwater demand is high. The analyses do not consider subsidence magnitude and rate, owing to the lack of this information at a global scale. Hence, the combination of subsidence susceptibility and the probability of groundwater depletion is used to predict a “proxy” of subsidence hazard, which permits identification of exposed areas where the probability of land subsidence occurrence is high or very high. Even though these results do not necessarily translate to direct impacts or damages, they are useful for identifying potential subsidence areas where further local-scale analysis is necessary. T he comparison of our model predictions with an independent validation dataset reveals a 94% capability to distinguish between subsidence and nonsubsidence areas, according to the value of the area under the receiver operating characteristic curve (see SM). The global exposure to potential subsidence is evaluated by calculating the number of inhabitants living in potential subsidence areas, i.e., subsidence hazard proxy, and the equivalent gross domestic product (GDP). T his “proxy” of exposed assets is calculated assuming that GDP per capita is homogeneous within each country. Finally, the evolution of potential global subsidence and the related exposure is predicted for 2040 for a global change scenario based on steady population growth and increasing greenhouse gas emissions (Shared Socioeconomic Pathways 2, Representative Concentration Pathway 8.5), which accounts for the greatest sea-level rise projections. ![Figure][5] Potential global subsidence The color scale indicates the probability intervals classified from very low (VL) to very high (VH), for every 30-arcsec resolution pixel (1 km by 1 km at the Equator). The white hatched polygons indicate countries where groundwater data is unavailable, and the potential subsidence only includes information on the susceptibility. See maps of other regions in supplementary materials. GRAPHIC: N. DESAI/ SCIENCE Our results suggest that potential subsidence threatens 12 million km2 (8%) of the global land surface with a probability greater than 50% (MH to VH in the figure). Potential subsidence areas are concentrated in and near densely urban and irrigated areas with high water stress and high groundwater demand, overlying some of the largest and most depleted aquifer systems ([ 5 ][6]) in Asia (e.g., North China Plain) and North America (e.g., Gulf of Mexico coastal plain); coastal and river delta areas worldwide (e.g., Vietnam, Egypt, or the Netherlands); and inland sedimentary basins of México, Iran, and the Mediterranean countries. Potential subsidence is lower in Africa, Australia, and South America, owing to the lower groundwater depletion ([ 6 ][7]). In central Africa, potential subsidence only includes information on the susceptibility, as groundwater depletion is unknown. In this region, subsidence susceptibility (see fig. S6) could be useful to prevent subsidence impacts on developing cities that during the next decades could rely more on the available groundwater resources. To evaluate the exposure to potential subsidence, we focus on areas where the potential subsidence probability is high or very high (see the figure). The cumulative potential subsidence area amounts to 2.2 million km2, or 1.6% of the land; includes 1.2 billion inhabitants, or 19% of the global population; and has an exposed GDP of US$ 8.19 trillion, or 12% of the global GDP. Hi gh-income countries account for 62% of the global exposed GDP but only 11% of the global exposed population, whereas low-income countries account for 54% of the global exposed population and 12% of the global exposed GDP. It is expected that the capability of low-income countries to implement the political, regulatory, and socioeconomic measures necessary to prevent and mitigate subsidence impact will be less than that for high-income countries. Potential subsidence threatens 484 million inhabitants living in flood-prone areas, 75% of whom live in fluvial areas and 25% of whom live near the coast. This number of threatened inhabitants corresponds to 50% of the global population exposed to flooding hazards according to previous estimates ([ 7 ][8]), demonstrating the importance of considering potential subsidence in global flooding risk analyses. Most of the global population exposed to potential subsidence live in Asia (86%), which is about 10 times the combined exposed population of North America and Europe (9%). The results indicate that 97% of the exposed global population is concentrated in 30 countries (see SM). India and China share the top two rankings of potential subsidence in terms of spatial extent and exposed population. Egypt and the Netherlands have the largest populations living in potential subsidence areas that are below the mean sea level. The greatest population densities in potential subsidence areas occur in Egypt and Indonesia, whereas the relative exposure per country, measured as the exposed population normalized by the total population, is greater than 30% for Egypt, Bangladesh, Netherlands, and Italy. The United States ranks first in terms of GDP exposed to potential subsidence, owing to its high GDP per capita. Combination of the aforementioned metrics permits derivation of a potential subsidence index ranking (see SM). Seven of the first ten ranked countries have the greatest subsidence impact, accounting for the greatest amount of reported damages (Netherlands, China, USA, Japan, Indonesia, México and Italy). During this century, climate change will cause serious impacts on the world's water resources through sea-level rise, more frequent and severe floods and droughts, changes in the mean value and mode of precipitation (rain versus snow), and increased evapotranspiration. Prolonged droughts will decrease groundwater recharge and increase groundwater depletion, intensifying subsidence. The global potential subsidence is predicted for 2040 using the same subsidence metrics and available global projections of water stress, water demand variations, climate, and population (see SM). Although predicted potential subsidence areas increase only by 7% globally, the threatened population is predicted to rise by 30%, affecting 1.6 billion inhabitants, 635 million of whom will be living in flood-prone areas. These changes will not be homogeneous. Between 2010 and 2040, the predicted population exposed to potential subsidence increases more than 80% in the Philippines, Iraq, Indonesia, México, Israel, Netherlands, Algeria, and Bangladesh. The increase will be moderate, less than 30%, for China, the United States, Italy, and Iran. Potential subsidence is forecasted to decrease in Japan and Germany, owing to effective groundwater management policies and population declines. Finally, potential subsidence is predicted to emerge in high-latitude northern countries like Canada and to increase in extent in Russia or Hungary, where climate change will favor longer dry seasons. Further advancements in the global evaluation of subsidence can be made when a global historical database on subsidence rate, magnitude, and extent has been compiled, which could be largely sourced from continental monitoring of surface displacements using satellite radar imagery ([ 8 ][9]). Widespread continuous monitoring of subsidence will permit better evaluation of the potential impact of land subsidence, especially in countries like Indonesia, México, and Iran, where local studies revealed the highest subsidence rates worldwide, but the national dimension of subsidence is still unknown. Further research also is necessary to evaluate the cost of damage caused by current and historical subsidence worldwide. The combination of damage information with hazard estimates will permit improved assessments of potential loss and design of cost-effective countermeasures. Presently, annual subsidence costs are only published for China (US$ 1.5 billion) and the Netherlands (US$ 4.8 billion) ([ 9 ][10]). The greater subsidence costs in the Netherlands owe to the exposed population below the mean sea level and the large investments made to prevent flooding. Our model, which does not yet consider mitigation measures, likely overestimates potential subsidence exposure in the Netherlands and Japan, where groundwater management has effectively controlled subsidence over the past decades ([ 10 ][11]). Our results identify 1596 major cities, or about 22% of the world's 7343 major cities that are in potential subsidence areas, with 57% of these cities also located in flood-prone areas. Moreover, subsidence threatens 15 of the 20 major coastal cities ranked with the highest flood risk worldwide ([ 11 ][12]), where potential subsidence can help delimit areas in which flooding risk could be increased and mitigation measures are necessary. Overall, potential global subsidence results can be useful to better define the spatial extent of poorly documented subsidence occurrences, discover unknown subsiding areas, prevent potential subsidence impacts wherever groundwater depletion occurs, and better identify areas where subsidence could increase the flooding risk. In any of these scenarios, an effective land-subsidence policy should include systematic monitoring and modeling of exposed areas, evaluation of potential damages, and cost-benefit analyses permitting implementation of adequate mitigation or adaptation measures. These measures should consider groundwater regulation and strategic long-term measures, such as the development of alternative water supplies and the protection and (or) enhancement of natural or artificial recharge of aquifers. Considering that the potential subsidence may affect 635 million inhabitants living in flood-prone areas in 2040, it is of prime importance that potential subsidence is quantified and systematically included in flood risk analyses and related mitigation strategies. [science.sciencemag.org/content/371/6524/34/suppl/DC1][13] 1. [↵][14]1. D. L. Galloway, 2. T. J. Burbey , Hydrogeol. J. 19, 1459 (2011). [OpenUrl][15] 2. [↵][16]1. J. S. Famiglietti , Nat. Clim. Chang. 4, 945 (2014). [OpenUrl][17] 3. [↵][18]1. K. E. Trenberth , Clim. Res. 47, 123 (2011). [OpenUrl][19][CrossRef][20][Web of Science][21] 4. [↵][22]1. J. P. M. Syvitski et al ., Nat. Geosci. 2, 681 (2009). [OpenUrl][23][CrossRef][24][Web of Science][25] 5. [↵][26]1. P. Döll, 2. H. Müller Schmied, 3. C. Schuh, 4. F. T. Portmann, 5. A. Eicker , Water Resour. Res. 50, 5698 (2014). [OpenUrl][27][CrossRef][28][PubMed][29] 6. [↵][30]1. R. G. Taylor et al ., Nat. Clim. Chang. 3, 322 (2013). [OpenUrl][31] 7. [↵][32]1. B. Jongman, 2. P. J. Ward, 3. J. C. J. H. Aerts , Glob. Environ. Change 22, 823 (2012). [OpenUrl][33] 8. [↵][34]1. R. Lanari et al ., Remote Sens. 12, 2961 (2020). [OpenUrl][35] 9. [↵][36]1. T. H. M. Bucx, 2. C. J. M. Van Ruiten, 3. G. Erkens, 4. G. De Lange , in Proceedings of the International Association of Hydrological Sciences 372, 485 (2015). [OpenUrl][37] 10. [↵][38]1. K. A. B. Jago-on et al ., Sci. Total Environ. 407, 3089 (2009). [OpenUrl][39][CrossRef][40][PubMed][41] 11. [↵][42]1. S. Hallegatte, 2. C. Green, 3. R. J. Nicholls, 4. J. Corfee-Morlot , Nat. Clim. Chang. 3, 802 (2013). [OpenUrl][43] 12. [↵][44]1. G. Herrera, 2. P. Ezquerro , Global Subsidence Maps, figshare (2020); 10.6084/m9.figshare.13312070. Acknowledgments: Four anonymous peer reviewers and S. E. Ingebritsen (U.S. Geological Survey) helped to improve the manuscript. Funding for this study was provided partly by the Spanish Research Agency (AQUARISK, PRX19/00065, TEC2017-85244-C2-1-P projects) and PRIMA RESERVOIR project, and by all the institutions represented in the Land Subsidence International Initiative from UNESCO. G.H.-G., P.E., R.T., M.B.-P, and J.L.-V. designed the study, performed the analysis, and wrote the initial manuscript with input from all other authors. R.M.M., E.C.-C., and M.R. advised on the susceptibility analysis. R.M.M., J.L., P.T., and G.E. advised on hazard analysis. D.C.-F., J.L., P.T., E.C.C., G.E., D.G., W.C.H., N.K., M.S., L.T., H.W., and S.Y. advised on global exposure analysis. R.T., M.B.P., R.M.M., J.L., P.T., W.-C.H., N.K., L.T., H.W., and S.Y. contributed essential data for the analysis. All the authors edited and revised the manuscript through the different reviews. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. government. The authors declare no competing interests. All data included in this study are available at figshare ([ 12 ][45]). [1]: #ref-1 [2]: #ref-2 [3]: #ref-3 [4]: #ref-4 [5]: pending:yes [6]: #ref-5 [7]: #ref-6 [8]: #ref-7 [9]: #ref-8 [10]: #ref-9 [11]: #ref-10 [12]: #ref-11 [13]: http://science.sciencemag.org/content/371/6524/34/suppl/DC1 [14]: #xref-ref-1-1 "View reference 1 in text" [15]: {openurl}?query=rft.jtitle%253DHydrogeol.%2BJ.%26rft.volume%253D19%26rft.spage%253D1459%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [16]: #xref-ref-2-1 "View reference 2 in text" [17]: {openurl}?query=rft.jtitle%253DNat.%2BClim.%2BChang.%26rft.volume%253D4%26rft.spage%253D945%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [18]: #xref-ref-3-1 "View reference 3 in text" [19]: {openurl}?query=rft.jtitle%253DClim.%2BRes.%26rft.volume%253D47%26rft.spage%253D123%26rft_id%253Dinfo%253Adoi%252F10.3354%252Fcr00953%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [20]: /lookup/external-ref?access_num=10.3354/cr00953&link_type=DOI [21]: /lookup/external-ref?access_num=000289207700014&link_type=ISI [22]: #xref-ref-4-1 "View reference 4 in text" [23]: {openurl}?query=rft.jtitle%253DNat.%2BGeosci.%26rft.volume%253D2%26rft.spage%253D681%26rft_id%253Dinfo%253Adoi%252F10.1038%252Fngeo629%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [24]: /lookup/external-ref?access_num=10.1038/ngeo629&link_type=DOI [25]: /lookup/external-ref?access_num=000270355000011&link_type=ISI [26]: #xref-ref-5-1 "View reference 5 in text" [27]: {openurl}?query=rft.jtitle%253DWater%2BResour.%2BRes.%26rft.volume%253D50%26rft.spage%253D5698%26rft_id%253Dinfo%253Adoi%252F10.1002%252F2014WR015595%26rft_id%253Dinfo%253Apmid%252F32020988%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [28]: /lookup/external-ref?access_num=10.1002/2014WR015595&link_type=DOI [29]: /lookup/external-ref?access_num=32020988&link_type=MED&atom=%2Fsci%2F371%2F6524%2F34.atom [30]: #xref-ref-6-1 "View reference 6 in text" [31]: {openurl}?query=rft.jtitle%253DNat.%2BClim.%2BChang.%26rft.volume%253D3%26rft.spage%253D322%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [32]: #xref-ref-7-1 "View reference 7 in text" [33]: {openurl}?query=rft.jtitle%253DGlob.%2BEnviron.%2BChange%26rft.volume%253D22%26rft.spage%253D823%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [34]: #xref-ref-8-1 "View reference 8 in text" [35]: {openurl}?query=rft.jtitle%253DRemote%2BSens.%26rft.volume%253D12%26rft.spage%253D2961%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [36]: #xref-ref-9-1 "View reference 9 in text" [37]: {openurl}?query=rft.jtitle%253DProceedings%2Bof%2Bthe%2BInternational%2BAssociation%2Bof%2BHydrological%2BSciences%26rft.volume%253D372%26rft.spage%253D485%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [38]: #xref-ref-10-1 "View reference 10 in text" [39]: {openurl}?query=rft.jtitle%253DThe%2BScience%2Bof%2Bthe%2BTotal%2BEnvironment%26rft.stitle%253DThe%2BScience%2Bof%2Bthe%2BTotal%2BEnvironment%26rft.aulast%253DJago-On%26rft.auinit1%253DK.%2BA.%26rft.volume%253D407%26rft.issue%253D9%26rft.spage%253D3089%26rft.epage%253D3104%26rft.atitle%253DUrbanization%2Band%2Bsubsurface%2Benvironmental%2Bissues%253A%2Ban%2Battempt%2Bat%2BDPSIR%2Bmodel%2Bapplication%2Bin%2BAsian%2Bcities.%26rft_id%253Dinfo%253Adoi%252F10.1016%252Fj.scitotenv.2008.08.004%26rft_id%253Dinfo%253Apmid%252F18922564%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [40]: /lookup/external-ref?access_num=10.1016/j.scitotenv.2008.08.004&link_type=DOI [41]: /lookup/external-ref?access_num=18922564&link_type=MED&atom=%2Fsci%2F371%2F6524%2F34.atom [42]: #xref-ref-11-1 "View reference 11 in text" [43]: {openurl}?query=rft.jtitle%253DNat.%2BClim.%2BChang.%26rft.volume%253D3%26rft.spage%253D802%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [44]: #xref-ref-12-1 "View reference 12 in text" [45]: #ref-12

New York (CNN Business)Talk of a possible Apple car is back. Apple (AAPL) hasn't commented publicly on its plans for the project, nicknamed Titan, so it's not clear exactly what will come of the effort. Some who follow the company think it could release a whole Apple-branded, electric, self-driving car. Others think it's more likely Apple will partner with existing automakers to sell an operating system (iDrive, maybe?), self-driving tools or other technology. There are some clues available, though.

While routing in wireless networks has been studied extensively, existing protocols are typically designed for a specific set of network conditions and so cannot accommodate any drastic changes in those conditions. For instance, protocols designed for connected networks cannot be easily applied to disconnected networks. In this paper, we develop a distributed routing strategy based on deep reinforcement learning that generalizes to diverse traffic patterns, congestion levels, network connectivity, and link dynamics. We make the following key innovations in our design: (i) the use of relational features as inputs to the deep neural network approximating the decision space, which enables our algorithm to generalize to diverse network conditions, (ii) the use of packet-centric decisions to transform the routing problem into an episodic task by viewing packets, rather than wireless devices, as reinforcement learning agents, which provides a natural way to propagate and model rewards accurately during learning, and (iii) the use of extended-time actions to model the time spent by a packet waiting in a queue, which reduces the amount of training data needed and allows the learning algorithm to converge more quickly. We evaluate our routing algorithm using a packet-level simulator and show that the policy our algorithm learns during training is able to generalize to larger and more congested networks, different topologies, and diverse link dynamics. Our algorithm outperforms shortest path and backpressure routing with respect to packets delivered and delay per packet.

We review the EfficientQA competition from NeurIPS 2020. The competition focused on open-domain question answering (QA), where systems take natural language questions as input and return natural language answers. The aim of the competition was to build systems that can predict correct answers while also satisfying strict on-disk memory budgets. These memory budgets were designed to encourage contestants to explore the trade-off between storing large, redundant, retrieval corpora or the parameters of large learned models. In this report, we describe the motivation and organization of the competition, review the best submissions, and analyze system predictions to inform a discussion of evaluation for open-domain QA.

This story originally appeared on WIRED UK. This is not a year we'll look back on fondly. It began with Australia on fire and ends with more than 1.5 million dead in a pandemic. But there have been bright points in this annus horribilis. While many of us saved lives by hunkering down at home watching Netflix, a communal act of selflessness that shouldn't be soon forgotten, progress was made in science, the environment, and even politics--Biden won!

Deep learning methods are achieving ever-increasing performance on many artificial intelligence tasks. A major limitation of deep models is that they are not amenable to interpretability. This limitation can be circumvented by developing post hoc techniques to explain the predictions, giving rise to the area of explainability. Recently, explainability of deep models on images and texts has achieved significant progress. In the area of graph data, graph neural networks (GNNs) and their explainability are experiencing rapid developments. However, there is neither a unified treatment of GNN explainability methods, nor a standard benchmark and testbed for evaluations. In this survey, we provide a unified and taxonomic view of current GNN explainability methods. Our unified and taxonomic treatments of this subject shed lights on the commonalities and differences of existing methods and set the stage for further methodological developments. To facilitate evaluations, we generate a set of benchmark graph datasets specifically for GNN explainability. We summarize current datasets and metrics for evaluating GNN explainability. Altogether, this work provides a unified methodological treatment of GNN explainability and a standardized testbed for evaluations.

Gaussian processes (GPs) are a popular model for spatially referenced data and allow descriptive statements, predictions at new locations, and simulation of new fields. Often a few parameters are sufficient to parameterize the covariance function, and maximum likelihood (ML) methods can be used to estimate these parameters from data. ML methods, however, are computationally demanding. For example, in the case of local likelihood estimation, even fitting covariance models on modest size windows can overwhelm typical computational resources for data analysis. This limitation motivates the idea of using neural network (NN) methods to approximate ML estimates. We train NNs to take moderate size spatial fields or variograms as input and return the range and noise-to-signal covariance parameters. Once trained, the NNs provide estimates with a similar accuracy compared to ML estimation and at a speedup by a factor of 100 or more. Although we focus on a specific covariance estimation problem motivated by a climate science application, this work can be easily extended to other, more complex, spatial problems and provides a proof-of-concept for this use of machine learning in computational statistics.

An abduction-based diagnosis technique for a class of discrete-event systems (DESs), called deep DESs (DDESs), is presented. A DDES has a tree structure, where each node is a network of communicating automata, called an active unit (AU). The interaction of components within an AU gives rise to emergent events. An emergent event occurs when specific components collectively perform a sequence of transitions matching a given regular language. Any event emerging in an AU triggers the transition of a component in its parent AU. We say that the DDES has a deep behavior, in the sense that the behavior of an AU is governed not only by the events exchanged by the components within the AU but also by the events emerging from child AUs. Deep behavior characterizes not only living beings, including humans, but also artifacts, such as robots that operate in contexts at varying abstraction levels. Surprisingly, experimental results indicate that the hierarchical complexity of the system translates into a decreased computational complexity of the diagnosis task. Hence, the diagnosis technique is shown to be (formally) correct as well as (empirically) efficient.

A generalist robot must be able to complete a variety of tasks in its environment. One appealing way to specify each task is in terms of a goal observation. However, learning goal-reaching policies with reinforcement learning remains a challenging problem, particularly when hand-engineered reward functions are not available. Learned dynamics models are a promising approach for learning about the environment without rewards or task-directed data, but planning to reach goals with such a model requires a notion of functional similarity between observations and goal states. We present a self-supervised method for model-based visual goal reaching, which uses both a visual dynamics model as well as a dynamical distance function learned using model-free reinforcement learning. Our approach learns entirely using offline, unlabeled data, making it practical to scale to large and diverse datasets. In our experiments, we find that our method can successfully learn models that perform a variety of tasks at test-time, moving objects amid distractors with a simulated robotic arm and even learning to open and close a drawer using a real-world robot. In comparisons, we find that this approach substantially outperforms both model-free and model-based prior methods. Videos and visualizations are available here: https://sites.google.com/berkeley.edu/mbold. Designing general-purpose robots that can perform a wide range of tasks remains an open problem in AI and robotics. Reinforcement learning (RL) represents a particularly promising tool for learning robotic behaviors when skills can be learned one at a time from user-defined reward functions. However, general-purpose robots will likely require large and diverse repertoires of skills, and learning individual tasks one at a time from manually-specified rewards is onerous and time-consuming. How can we design learning systems that can autonomously acquire general-purpose knowledge that allows them to solve many different downstream tasks? To address this problem, we must resolve three questions.

Nowadays it is inevitable to use intelligent systems to improve the performance and optimization of different components of devices or factories. Furthermore, it's so essential to have appropriate predictions to make better decisions in businesses, medical studies, and engineering studies, etc. One of the newest and most widely used of these methods is a field called'Data Science' that all of the scientists, engineers, and factories need to learn and use them in their careers. This article briefly introduced data science and reviewed its methods, especially it's usages in mechanical engineering and challenges and ways of developing data science in mechanical engineering. In the introduction, different definitions of data science and its background in technology reviewed. In the following, data science methodology which is the process that a data scientist needs to do in its works been discussed. Further, some researches in mechanical engineering area that used data science methods in their studies, are reviewed. Eventually, it has been discussed according to the subjects that have been reviewed in the article, why it is necessary to use data science in mechanical engineering researches and projects.