An official website of the United States government
Abstract Researchers use both experiments and observations to study the impacts of climate change on ecosystems, but results from these contrasting approaches have not been systematically compared for droughts. Using a meta-analysis and accounting for potential confounding factors, we demonstrate that aboveground biomass responded only about half as much to experimentally imposed drought events as to natural droughts. Our findings indicate that experimental results may underestimate climate change impacts and highlight the need to integrate results across approaches.
more »
« less
Replication Data for: Experimental Evaluation of Algorithm-Assisted Human Decision-Making: Application to Pretrial Public Safety Assessment
aihuman is an R package which provides statistical methods for analyzing experimental evaluation of the causal impacts of algorithmic recommendations on human decisions developed by Imai, Jiang, Greiner, Halen, and Shin (2023) . The data used for this paper, and made available here, are interim, based on only half of the observations in the study and (for those observations) only half of the study follow-up period. We use them only to illustrate methods, not to draw substantive conclusions.
more »
« less
- Award ID(s):
- 2051196
- PAR ID:
- 10514520
- Publisher / Repository:
- Harvard Dataverse
- Date Published:
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Humans excel at efficiently navigating through crowds without collision by focusing on specific visual regions relevant to navigation. However, most robotic visual navigation methods rely on deep learning models pre-trained on vision tasks, which prioritize salient objects—not necessarily relevant to navigation and potentially misleading. Alternative approaches train specialized navigation models from scratch, requiring significant computation. On the other hand, self-supervised learning has revolutionized computer vision and natural language processing, but its application to robotic navigation remains underexplored due to the difficulty of defining effective self-supervision signals. Motivated by these observations, in this work, we propose a Self-Supervised Vision-Action Model for Visual Navigation Pre-Training (VANP). Instead of detecting salient objects that are beneficial for tasks such as classification or detection, VANP learns to focus only on specific visual regions that are relevant to the navigation task. To achieve this, VANP uses a history of visual observations, future actions, and a goal image for self-supervision, and embeds them using two small Transformer Encoders. Then, VANP maximizes the information between the embeddings by using a mutual information maximization objective function. We demonstrate that most VANP-extracted features match with human navigation intuition. VANP achieves comparable performance as models learned end-to-end with half the training time and models trained on a large-scale, fully supervised dataset, i.e., ImageNet, with only 0.08% data.more » « less
-
ABSTRACT The observationally inferred size versus stellar–mass relationship (SMR) for low-mass galaxies provides an important test for galaxy formation models. However, the relationship relies on assumptions that relate observed luminosity profiles to underlying stellar mass profiles. Here we use the Feedback in Realistic Environments simulations of low-mass galaxies to explore how the predicted SMR changes depending on whether one uses star-particle counts directly or mock observations. We reproduce the SMR found in The Exploration of Local Volume Satellites survey remarkably well only when we infer stellar masses and sizes using mock observations. However, when we use star particles to directly infer stellar masses and half-mass radii, we find that our galaxies are too large and obey an SMR with too little scatter compared to observations. This discrepancy between the ‘true’ galaxy size and mass and those derived in the mock observation approach is twofold. First, our simulated galaxies have higher and more varied mass-to-light ratios (MLR) at a fixed colour than those commonly adopted, which tends to underestimate their stellar masses compared to their true, simulated values. Second, our galaxies have radially increasing MLR gradients therefore using a single MLR tends to underpredict the mass in the outer regions. Similarly, the true half-mass radius is larger than the half-light radius because the light is more concentrated than the mass. If our simulations are accurate representations of the real Universe, then the relationship between galaxy size and stellar mass is even tighter for low-mass galaxies than is commonly inferred from observed relations.more » « less
-
null (Ed.)Abstract The leading interannual mode of winter surface air temperature over the North American (NA) sector, characterized by a “Warm Arctic, Cold Continents” (WACC) pattern, exerts pronounced influences on NA weather and climate, while its underlying mechanisms remain elusive. In this study, the relative roles of surface boundary forcing versus internal atmospheric processes for the formation of the WACC pattern are quantitatively investigated using a combined analysis of observations and large-ensemble atmospheric global climate model simulations. Internal atmospheric variability is found to play an important role in shaping the year-to-year WACC variability, contributing to about half of the total variance. An anomalous SST pattern resembling the North Pacific Mode is identified as a major surface boundary forcing pattern in driving the interannual WACC variability over the NA sector, with a minor contribution from sea ice variability over the Chukchi- Bering Seas. Findings from this study not only lead to improved understanding of underlying physics regulating the interannual WACC variability, but also provide important guidance for improved modeling and prediction of regional climate variability over NA and the Arctic region.more » « less
-
We present an efficient, trivially parallelizable algorithm to compute offset surfaces of shapes discretized using a dexel data structure. Our algorithm is based on a two-stage sweeping procedure that is simple to implement and efficient, entirely avoiding volumetric distance field computations typical of existing methods. Our construction is based on properties of half-space power diagrams, where each seed is only visible by a half space, which were never used before for the computation of surface offsets. The primary application of our method is interactive modeling for digital fabrication. Our technique enables a user to interactively process high-resolution models. It is also useful in a plethora of other geometry processing tasks requiring fast, approximate offsets, such as topology optimization, collision detection, and skeleton extraction. We present experimental timings, comparisons with previous approaches, and provide a reference implementation in the supplemental material.more » « less
