Edgar Josafat Martinez-Noriega’s research while affiliated with National Institute of Advanced Industrial Science and Technology and other places

Tropospheric ozone formation depends on the emissions of volatile organic compounds (VOC) and nitrogen oxides (NOx). In megacities, abundant VOC and NOx sources cause relentlessly high ozone episodes, affecting a large share of the global population. This study uses data from the Ozone Monitoring Instrument for formaldehyde (HCHO) and nitrogen dioxide (NO2) as proxy data for VOC and NOx emissions, respectively, with their ratio serving as an indicator of ozone sensitivity. Ground-level ozone (O3) reanalysis from the Copernicus Atmosphere Monitoring is used to assess the O3 trends. We evaluate changes from 2005 to 2019 and their relationship with the warming environment in 41 megacities worldwide, applying seasonal Mann-Kendall, trend decomposition methods, and Pearson correlation analysis. We reveal significant increases in global HCHO (0.1 to 0.31 × 10¹⁵ mol cm⁻² year⁻¹) and regionally varying NO2 (−0.22 to 0.07 × 10¹⁵ mol cm⁻² year⁻¹). O3 trends range from −0.31 to 0.70 ppb year⁻¹, highlighting the relevance of precursor abundance on O3 levels. The strong correlation between precursor emissions and increasing temperature suggests that O3 will continue to rise as climate change persists.

In the present work, we show that the performance of formula-driven supervised learning (FDSL) can match or even exceed that of ImageNet-21k without the use of real images, human-, and self-supervision during the pre-training of Vision Transformers (ViTs). For example, ViT-Base pre-trained on ImageNet-21k shows 81.8% top-1 accuracy when fine-tuned on ImageNet-1k and FDSL shows 82.7% top-1 accuracy when pre-trained under the same conditions (number of images, hyperparameters, and number of epochs). Images generated by formulas avoid the privacy/copyright issues, labeling cost and errors, and biases that real images suffer from, and thus have tremendous potential for pre-training general models. To understand the performance of the synthetic images, we tested two hypotheses, namely (i) object contours are what matter in FDSL datasets and (ii) increased number of parameters to create labels affects performance improvement in FDSL pre-training. To test the former hypothesis, we constructed a dataset that consisted of simple object contour combinations. We found that this dataset can match the performance of fractals. For the latter hypothesis, we found that increasing the difficulty of the pre-training task generally leads to better fine-tuning accuracy.