Search for: All records

How do practitioners who develop consumer AI products scope, motivate, and conduct privacy work? Respecting pri- vacy is a key principle for developing ethical, human-centered AI systems, but we cannot hope to better support practitioners without answers to that question. We interviewed 35 industry AI practitioners to bridge that gap. We found that practitioners viewed privacy as actions taken against pre-defined intrusions that can be exacerbated by the capabilities and requirements of AI, but few were aware of AI-specific privacy intrusions documented in prior literature. We found that their privacy work was rigidly defined and situated, guided by compliance with privacy regulations and policies, and generally demoti- vated beyond meeting minimum requirements. Finally, we found that the methods, tools, and resources they used in their privacy work generally did not help address the unique pri- vacy risks introduced or exacerbated by their use of AI in their products. Collectively, these findings reveal the need and opportunity to create tools, resources, and support structures to improve practitioners’ awareness of AI-specific privacy risks, motivations to do AI privacy work, and ability to ad- dress privacy harms introduced or exacerbated by their use of AI in consumer products. 

Privacy is a key principle for developing ethical AI technologies, but how does including AI technologies in products and services change privacy risks? We constructed a taxonomy of AI privacy risks by an- alyzing 321 documented AI privacy incidents. We codifed how the unique capabilities and requirements of AI technologies described in those incidents generated new privacy risks, exacerbated known ones, or otherwise did not meaningfully alter the risk. We present 12 high-level privacy risks that AI technologies either newly created (e.g., exposure risks from deepfake pornography) or exacerbated (e.g., surveillance risks from collecting training data). One upshot of our work is that incorporating AI technologies into a product can alter the privacy risks it entails. Yet, current approaches to privacy-preserving AI/ML (e.g., federated learning, diferential pri- vacy, checklists) only address a subset of the privacy risks arising from the capabilities and data requirements of AI. 

Much of our knowledge of the North American lithosphere comes from imaging seismic velocities. Additional constraints on the subsurface can be gained by studying seismic attenuation, which has different sensitivity to physical properties. We produce a model of lateral variations in attenuation across the conterminous U.S. by analyzing data recorded by the EarthScope Transportable Array. We divide the study area into 12 overlapping tiles and differential attenuation is measured in each tile independently; and twice for four of the tiles. Measurements are combined into a smooth map using a set of linear inversions. Comparing results for adjacent tiles and for repeated tiles shows that the imaged features are robust. The final map shows generally higher attenuation west of the Rocky Mountain Front than east of it, with significant small length scale variations superimposed on that broad pattern. In general, there is a strong anticorrelation between differential attenuation and shear wave velocities at depths of 80–250 km. However, a given change in velocity may correspond to a large or small change in attenuation, depending on the area; suggesting that different physical mechanisms are operating. In the western and south‐central U.S., as well as the Appalachians, velocity variations are large compared to attenuation changes, while the opposite is true in the north‐central and southeastern U.S. Calculations with the Very Broadband Rheology calculator show that these results are consistent with the main source of heterogeneity being temperature and melt fraction in the former regions and grain size variability in the latter ones.

 

Score matching is an alternative to maximum likelihood (ML) for estimating a probability distribution parametrized up to a constant of proportionality. By fitting the ''score'' of the distribution, it sidesteps the need to compute this constant of proportionality (which is often intractable). While score matching and variants thereof are popular in practice, precise theoretical understanding of the benefits and tradeoffs with maximum likelihood---both computational and statistical---are not well understood. In this work, we give the first example of a natural exponential family of distributions such that the score matching loss is computationally efficient to optimize, and has a comparable statistical efficiency to ML, while the ML loss is intractable to optimize using a gradient-based method. The family consists of exponentials of polynomials of fixed degree, and our result can be viewed as a continuous analogue of recent developments in the discrete setting. Precisely, we show: (1) Designing a zeroth-order or first-order oracle for optimizing the maximum likelihood loss is NP-hard. (2) Maximum likelihood has a statistical efficiency polynomial in the ambient dimension and the radius of the parameters of the family. (3) Minimizing the score matching loss is both computationally and statistically efficient, with complexity polynomial in the ambient dimension. 

In this study, 82 middle and high school teachers engaged with the InSTEP online professional learning platform to develop their expertise in teaching data science and statistics. We investigated teachers’ engagement within the platform, aspects of the platform that were most and least effective in building teachers’ expertise, and the extent to which teachers’ self-efficacy changed. Using mixed methods, we collected, analyzed and integrated multiple data sources. 

Shock–bubble interactions (SBIs) are important across a wide range of physical systems. In inertial confinement fusion, interactions between laser-driven shocks and micro-voids in both ablators and foam targets generate instabilities that are a major obstacle in achieving ignition. Experiments imaging the collapse of such voids at high energy densities (HED) are constrained by spatial and temporal resolution, making simulations a vital tool in understanding these systems. In this study, we benchmark several radiation and thermal transport models in the xRAGE hydrodynamic code against experimental images of a collapsing mesoscale void during the passage of a 300 GPa shock. We also quantitatively examine the role of transport physics in the evolution of the SBI. This allows us to understand the dynamics of the interaction at timescales shorter than experimental imaging framerates. We find that all radiation models examined reproduce empirical shock velocities within experimental error. Radiation transport is found to reduce shock pressures by providing an additional energy pathway in the ablation region, but this effect is small (∼1% of total shock pressure). Employing a flux-limited Spitzer model for heat conduction, we find that flux limiters between 0.03 and 0.10 produce agreement with experimental velocities, suggesting that the system is well-within the Spitzer regime. Higher heat conduction is found to lower temperatures in the ablated plasma and to prevent secondary shocks at the ablation front, resulting in weaker primary shocks. Finally, we confirm that the SBI-driven instabilities observed in the HED regime are baroclinically driven, as in the low energy case.

 

We report on a total of 310 samples from marine sediments drilled in the Indian Ocean that were analyzed for glass shard compositions. Samples are mainly from International Ocean Discovery Program Expeditions 353 and 362 but are complemented by samples from Expedition 354; Ocean Drilling Program Legs 183, 121, 120, 119, 116, and 115; and Deep Sea Drilling Project Leg 22. We performed 4327 successful single glass shard analyses with the electron microprobe for major element compositions and conducted 937 successful single analyses with laser ablation-inductively coupled plasma-mass spectrometry (LA-ICP-MS) for trace element compositions on individual glass shards previously measured with the electron microprobe. In total, we were able to measure glass compositions for 254 samples. Of all the samples, 235 can be classified as tephra layers containing pyroclasts as the predominant component in their clast inventory between the 63 and 125 µm grain size fraction, often exceeding 90 vol%. The compositions of the Indian Ocean marine tephras range from basalt to rhyolite and from basaltic trachyandesite to trachyte and fall into the calc-alkaline, K-rich calc-alkaline, and shoshonitic magmatic series.