A longstanding question in computer vision concerns the representation of 3D shapes for recognition: should 3D shapes be represented with descriptors operating on their native 3D formats, such as voxel grid or polygon mesh, or can they be effectively represented with view-based descriptors? We address this question in the context of learning to recognize 3D shapes from a collection of their rendered views on 2D images. We first present a standard CNN architecture trained to recognize the shapes' rendered views independently of each other, and show that a 3D shape can be recognized even from a single view at an accuracy far higher than using state-of-the-art 3D shape descriptors. Recognition rates further increase when multiple views of the shapes are provided. In addition, we present a novel CNN architecture that combines information from multiple views of a 3D shape into a single and compact shape descriptor offering even better recognition performance. The same architecture can be applied to accurately recognize human hand-drawn sketches of shapes. We conclude that a collection of 2D views can be highly informative for 3D shape recognition and is amenable to emerging CNN architectures and their derivatives.

Deep neural networks are vulnerable to adversarial examples, which poses security concerns on these algorithms due to the potentially severe consequences. Adversarial attacks serve as an important surrogate to evaluate the robustness of deep learning models before they are deployed. However, most of existing adversarial attacks can only fool a black-box model with a low success rate. To address this issue, we propose a broad class of momentum-based iterative algorithms to boost adversarial attacks. By integrating the momentum term into the iterative process for attacks, our methods can stabilize update directions and escape from poor local maxima during the iterations, resulting in more transferable adversarial examples. To further improve the success rates for black-box attacks, we apply momentum iterative algorithms to an ensemble of models, and show that the adversarially trained models with a strong defense ability are also vulnerable to our black-box attacks. We hope that the proposed methods will serve as a benchmark for evaluating the robustness of various deep models and defense methods. With this method, we won the first places in NIPS 2017 Non-targeted Adversarial Attack and Targeted Adversarial Attack competitions. * Corresponding author. Alps: 94.39%Dog: 99.99%Puffer: 97.99% Crab: 100.00%

Multiphase chemistry of NO2 and alkaline matter in aerosol water explains rapid sulfate formation and severe haze in China.

The MIX inventory is developed for the years 2008 and 2010 to support the Model Inter-Comparison Study for Asia (MICS-Asia) and the Task Force on Hemispheric Transport of Air Pollution (TF HTAP) by a mosaic of up-to-date regional emission inventories. Emissions are estimated for all major anthropogenic sources in 29 countries and regions in Asia. We conducted detailed comparisons of different regional emission inventories and incorporated the best available ones for each region into the mosaic inventory at a uniform spatial and temporal resolution. Emissions are aggregated to five anthropogenic sectors: power, industry, residential, transportation, and agriculture. We estimate the total Asian emissions of 10 species in 2010 as follows: 51.3 Tg SO2, 52.1 Tg NOx, 336.6 Tg CO, 67.0 Tg NMVOC (non-methane volatile organic compounds), 28.8 Tg NH3, 31.7 Tg PM10, 22.7 Tg PM2.5, 3.5 Tg BC, 8.3 Tg OC, and 17.3 Pg CO2. Emissions from China and India dominate the emissions of Asia for most of the species. We also estimated Asian emissions in 2006 using the same methodology of MIX. The relative change rates of Asian emissions for the period of 2006–2010 are estimated as follows: −8.1 % for SO2, +19.2 % for NOx, +3.9 % for CO, +15.5 % for NMVOC, +1.7 % for NH3, −3.4 % for PM10, −1.6 % for PM2.5, +5.5 % for BC, +1.8 % for OC, and +19.9 % for CO2. Model-ready speciated NMVOC emissions for SAPRC-99 and CB05 mechanisms were developed following a profile-assignment approach. Monthly gridded emissions at a spatial resolution of 0.25°  ×  0.25° are developed and can be accessed from http://www.meicmodel.org/dataset-mix

The Amazon is one of the few continental regions where atmospheric aerosol particles and their effects on climate are not dominated by anthropogenic sources. During the wet season, the ambient conditions approach those of the pristine pre-industrial era. We show that the fine submicrometer particles accounting for most cloud condensation nuclei are predominantly composed of secondary organic material formed by oxidation of gaseous biogenic precursors. Supermicrometer particles, which are relevant as ice nuclei, consist mostly of primary biological material directly released from rainforest biota. The Amazon Basin appears to be a biogeochemical reactor, in which the biosphere and atmospheric photochemistry produce nuclei for clouds and precipitation sustaining the hydrological cycle. The prevailing regime of aerosol-cloud interactions in this natural environment is distinctly different from polluted regions.

Extreme haze episodes repeatedly shrouded Beijing during the winter of 2012-2013, causing major environmental and health problems. To better understand these extreme events, we performed a model-assisted analysis of the hourly observation data of PM2.5 and its major chemical compositions. The synthetic analysis shows that (1) the severe winter haze was driven by stable synoptic meteorological conditions over northeastern China, and not by an abrupt increase in anthropogenic emissions. (2) Secondary species, including organics, sulfate, nitrate, and ammonium, were the major constituents of PM2.5 during this period. (3) Due to the dimming effect of high loading of aerosol particles, gaseous oxidant concentrations decreased significantly, suggesting a reduced production of secondary aerosols through gas-phase reactions. Surprisingly, the observational data reveals an enhanced production rate of secondary aerosols, suggesting an important contribution from other formation pathways, most likely heterogeneous reactions. These reactions appeared to be more efficient in producing secondary inorganics aerosols than organic aerosols resulting in a strongly elevated fraction of inorganics during heavily polluted periods. (4) Moreover, we found that high aerosol concentration was a regional phenomenon. The accumulation process of aerosol particles occurred successively from cities southeast of Beijing. The apparent sharp increase in PM2.5 concentration of up to several hundred mu g m(-3) per hour recorded in Beijing represented rapid recovery from an interruption to the continuous pollution accumulation over the region, rather than purely local chemical production. This suggests that regional transport of pollutants played an important role during these severe pollution events

Hydroxyl radicals (OH) are a key species in atmospheric photochemistry. In the lower atmosphere, up to ~30% of the primary OH radical production is attributed to the photolysis of nitrous acid (HONO), and field observations suggest a large missing source of HONO. We show that soil nitrite can release HONO and explain the reported strength and diurnal variation of the missing source. Fertilized soils with low pH appear to be particularly strong sources of HONO and OH. Thus, agricultural activities and land-use changes may strongly influence the oxidizing capacity of the atmosphere. Because of the widespread occurrence of nitrite-producing microbes, the release of HONO from soil may also be important in natural environments, including forests and boreal regions.

Aerosol‐planetary boundary layer (PBL) interactions have been found to enhance air pollution in megacities in China. We show that black carbon (BC) aerosols play the key role in modifying the PBL meteorology and hence enhancing the haze pollution. With model simulations and data analysis from various field observations in December 2013, we demonstrate that BC induces heating in the PBL, particularly in the upper PBL, and the resulting decreased surface heat flux substantially depresses the development of PBL and consequently enhances the occurrences of extreme haze pollution episodes. We define this process as the “dome effect” of BC and suggest an urgent need for reducing BC emissions as an efficient way to mitigate the extreme haze pollution in megacities of China.