More Like this

1. Detecting political biases of named entities and hashtags on Twitter

   https://doi.org/10.1140/epjds/s13688-023-00386-6  

   Xiao, Zhiping; Zhu, Jeffrey; Wang, Yining; Zhou, Pei; Lam, Wen Hong; Porter, Mason A.; Sun, Yizhou (June 2023, EPJ Data Science)

   Abstract Ideological divisions in the United States have become increasingly prominent in daily communication. Accordingly, there has been much research on political polarization, including many recent efforts that take a computational perspective. By detecting political biases in a text document, one can attempt to discern and describe its polarity. Intuitively, the named entities (i.e., the nouns and the phrases that act as nouns) and hashtags in text often carry information about political views. For example, people who use the term “pro-choice” are likely to be liberal and people who use the term “pro-life” are likely to be conservative. In this paper, we seek to reveal political polarities in social-media text data and to quantify these polarities by explicitly assigning a polarity score to entities and hashtags. Although this idea is straightforward, it is difficult to perform such inference in a trustworthy quantitative way. Key challenges include the small number of known labels, the continuous spectrum of political views, and the preservation of both a polarity score and a polarity-neutral semantic meaning in an embedding vector of words. To attempt to overcome these challenges, we propose thePolarity-awareEmbeddingMulti-task learning (PEM) model. This model consists of (1) a self-supervised context-preservation task, (2) an attention-based tweet-level polarity-inference task, and (3) an adversarial learning task that promotes independence between an embedding’s polarity component and its semantic component. Our experimental results demonstrate that ourPEMmodel can successfully learn polarity-aware embeddings that perform well at tweet-level and account-level classification tasks. We examine a variety of applications—including a study of spatial and temporal distributions of polarities and a comparison between tweets from Twitter and posts from Parler—and we thereby demonstrate the effectiveness of ourPEMmodel. We also discuss important limitations of our work and encourage caution when applying thePEMmodel to real-world scenarios. 

   more » « less
2. Earthquake focal mechanisms with distributed acoustic sensing

   https://doi.org/10.1038/s41467-023-39639-3  

   Li, Jiaxuan; Zhu, Weiqiang; Biondi, Ettore; Zhan, Zhongwen (July 2023, Nature Communications)

   Abstract Earthquake focal mechanisms provide critical in-situ insights about the subsurface faulting geometry and stress state. For frequent small earthquakes (magnitude< 3.5), their focal mechanisms are routinely determined using first-arrival polarities picked on the vertical component of seismometers. Nevertheless, their quality is usually limited by the azimuthal coverage of the local seismic network. The emerging distributed acoustic sensing (DAS) technology, which can convert pre-existing telecommunication cables into arrays of strain/strain-rate meters, can potentially fill the azimuthal gap and enhance constraints on the nodal plane orientation through its long sensing range and dense spatial sampling. However, determining first-arrival polarities on DAS is challenging due to its single-component sensing and low signal-to-noise ratio for direct body waves. Here, we present a data-driven method that measures P-wave polarities on a DAS array based on cross-correlations between earthquake pairs. We validate the inferred polarities using the regional network catalog on two DAS arrays, deployed in California and each comprising ~ 5000 channels. We demonstrate that a joint focal mechanism inversion combining conventional and DAS polarity picks improves the accuracy and reduces the uncertainty in the focal plane orientation. Our results highlight the significant potential of integrating DAS with conventional networks for investigating high-resolution earthquake source mechanisms. 

   more » « less
3. A Deep Learning-Aided Workflow for Decoding the Stress Regime of Southern Nevada

   https://doi.org/10.1785/0220240443  

   Chatterjee, Avigyan; Pennington, Colin N; Trugman, Daniel T; Walter, William R (March 2025, Seismological Research Letters)

   Abstract The Rock Valley fault zone in southern Nevada has a notable history of seismic activity and is the site of a future direct comparison experiment of explosion and earthquake sources. This study aims to gain insight into regional tectonic processes by leveraging recent advances in seismic monitoring capabilities to elucidate the local stress regime. A crucial step in this investigation is the accurate determination of P-wave first-motion polarities, which play a vital role in resolving earthquake focal mechanisms of small earthquakes. We deploy a deep learning-based method for automatic determination of first-motion polarities to vastly expand the polarity dataset beyond what has been reviewed by human analysts. By the integrating P-wave polarities with new measurements of S/P amplitude ratios, we obtain robust focal mechanism estimates for 1306 earthquakes with a local magnitude of 1 and above occurring between 2010 and 2023 in southern Nevada. We then use the focal mechanism catalog to examine the regional stress orientation, confirming an overall trans-tensional stress regime with smaller scale complexities illuminated by individual earthquake sequences. These findings demonstrate how detailed analyses of small earthquakes can provide fundamental information for understanding earthquake processes in the region and inform future experiments at the Nevada National Security Site. 

   more » « less
4. Deciphering Subduction Polarity During Ancient Arc‐Continent Collisions

   https://doi.org/10.1029/2024GL108761  

   Yan, Zhiyong; Chen, Lin; Zuza, Andrew V; Xiang, Xiao; Xie, Renxian; Ai, Sanxi (August 2024, Geophysical Research Letters)

   The closure of an ancient ocean basin via oceanic arc‐continent collision has two subduction styles with opposite polarities, which may proceed via subduction polarity reversal (SPR) or a subduction zone jump (SZJ). Interpreting the geometry or kinematic evolution of ancient collisional zones, especially the original subduction polarity, can be challenging. Here we used 2D thermo‐mechanical modeling to investigate the dynamic evolution process of SPR versus SZJ. Our modeling predicts different structural, topographic, magmatic, and basin histories for SPR and SZJ, which can be compared against, and help interpret, the geologic record past sites of oceanic closure during collisional orogens. Our results match geologic observations of past collisions in Kamchatka, eastern Russia, and the Banda Arc, eastern Indonesia, and thus our results can help effectively decode the evolutionary history of past arc‐continent collisions. 

   more » « less
5. SGD and Weight Decay Provably Induce a Low-Rank Bias in Deep Neural Networks

   Galanti, Tomer; Siegel, Zachary; Gupte, Aparna; Poggio, Tomaso (February 2023, Center for Brains, Minds and Machines (CBMM))

   In this paper, we study the bias of Stochastic Gradient Descent (SGD) to learn low-rank weight matrices when training deep ReLU neural networks. Our results show that training neural networks with mini-batch SGD and weight decay causes a bias towards rank minimization over the weight matrices. Specifically, we show, both theoretically and empirically, that this bias is more pronounced when using smaller batch sizes, higher learning rates, or increased weight decay. Additionally, we predict and observe empirically that weight decay is necessary to achieve this bias. Finally, we empirically investigate the connection between this bias and generalization, finding that it has a marginal effect on generalization. Our analysis is based on a minimal set of assumptions and applies to neural networks of any width or depth, including those with residual connections and convolutional layers. 

   more » « less