Although pollution is widespread, there is little evidence about how it might harm children’s long run outcomes. Using the detailed, geocoded data that follows national representative cohorts of children born to the National Longitudinal Survey of Youth respondents over time, I compare siblings who were gestating before versus after a Toxic Release Inventory site opened or closed within one mile of their home. I find that children who were exposed prenatally to industrial pollution have lower wages, are more likely to be in poverty as adults, have fewer years of completed education, and are less likely to graduate high school.Institutional subscribers to the NBER working paper series, and residents of developing countries may download this paper without additional charge at www.nber.org.
Chest X-rays are an inexpensive and widely available imaging modality for diagnosing or monitoring a variety of medical conditions. Given their abundance, healthcare providers could greatly benefit from automated systems capable of screening healthy patients and supporting the diagnosis of pathological cases. Deep learning has become central to such decision-support systems, offering accurate and efficient image classification that can improve clinical workflows and reduce radiologist workload. However, despite the rapid evolution of general-purpose neural architectures, particularly attention-based models, their application to medical imaging remains constrained by limited incorporation of medical domain knowledge. Most existing attention mechanisms optimize only task-specific losses, disregarding crucial anatomical and lesion-location priors, which can hinder generalization and interpretability. In this work, we introduce a fully automated, attention-guided classification framework that integrates medical priors through an on-the-fly segmentation of the lungs, followed by a spatially aware attention loss that directs the network’s focus toward clinically relevant regions. The method requires minimal physician input—only a single annotated X-ray indicating potential lesion areas at initialization and generalizes effectively across patients without relying on absolute bounding-box coordinates. Gradient-based activation mapping is further employed to ensure alignment between attention and lesion-specific regions. Our approach is architecture-agnostic and integrates seamlessly into end-to-end pipelines. Experiments on two medical image datasets demonstrate that the proposed segmentation-enhanced attention loss improves both classification accuracy and representation interpretability compared to the standard cross-entropy loss. The code is available at: https://github.com/rcorizzo/cxr-segmentation-attention/ .
Recent work in the structural approach to consciousness has shown great promise as a research paradigm for the formal and empirical study of the phenomenal qualities of experience, i.e., qualia. In this paradigm, qualia are characterized by modeling the internal organization of parts within an experience, or by modeling external relations between instances of experience. A major next step for the structural approach is to integrate these two perspectives into an account of phenomenally unified global experience. In this paper, we describe these two types of structural models and how their category-theoretic formalizations contribute to the task of identifying the physical bases of phenomenal consciousness. We then propose a sheaf-theoretic framework that integrates these two approaches by mapping mereological parts of experience to empirical measures of their qualia. Through an application to the experience of visual space, we demonstrate that this framework enables a formal description of the structure of experience and conditions for phenomenal unity. We discuss how this integrative approach supports an empirical research program for investigating the relationship between local and global phenomenal qualities, and outline directions for future work toward a structural characterization of global phenomenal consciousness.
The effective redshift distribution n(z) of galaxies is a critical component in the study of weak gravitational lensing. Here, we introduce a new method for determining n(z) for weak lensing surveys based on high-quality redshifts and neural-network-based importance weights. Additionally, we present the first unified photometric redshift calibration of the three leading stage-III weak lensing surveys, the Dark Energy Survey (DES), the Hyper Suprime-Cam (HSC) survey, and the Kilo-Degree Survey (KiDS), with state-of-the-art spectroscopic data from the Dark Energy Spectroscopic Instrument (DESI). We verify our method using a new, data-driven approach and obtain n(z) constraints with statistical uncertainties of the order of sigma z & strns;similar to 0.01 and smaller. Our analysis is largely independent of previous photometric redshift calibrations and, thus, provides an important cross-check in light of recent cosmological tensions. Overall, we find excellent agreement with previously published results on the DES Y3 and HSC Y1 data sets, while there are some differences on the mean redshift with respect to the previously published KiDS-1000 results. We attribute the latter to mismatches in photometric noise properties in the COSMOS field compared to the wider KiDS self-organizing map-gold catalog. At the same time, the new n(z) estimates for KiDS do not significantly change estimates of cosmic structure growth from cosmic shear. Finally, we discuss how our method can be applied to future weak lensing calibrations with DESI data.
