arXiv subjects

Xiyang Zhang

Publications and source records attributed to Xiyang Zhang.

18 recordsLinked to original sources

arXiv2026

ALIVE: Warnings Before Exclusion in Budgeted Multi-Source Learning

A routing decision can be revised at the next transaction, but a latched source exclusion persists across later decisions. We ask what evidence should authorize these unequal-persistence actions when finite-population auditing and learning share a budget. ALIVE (Action-Layered Intervention via Evidence) is an auditable control layer: one randomized without-replacement prefix supplies cached evidence, heuristic warnings drive non-latching floor-bounded routing, and only two fresh simultaneous certificate separations may latch an exclusion request subject to capacity-feasible activation. Conditional on fixed support and labels under an ideal uniform audit permutation, any predictable controller preserving this interface inherits an anytime familywise bound of δon acting against a source that fails the pre-fixed absolute or relative strict-majority-disagreement predicate. With a published known-size, all-strict-majority PPR engine, median evidence count fell from 304 to 96 identities in e40 and from 171 to 62 in e60, while both engines used 48 in e80. In the matched CIFAR controller, the persistent-action layer added +0.1935 accuracy-AUBC percentage points over routing-only in all ten paired seed clusters. The +0.1954-point full-system contrast against CBR was also positive but did not meet the predeclared multiplicity-adjusted criterion (conditional Holm-adjusted sign-flip reference value =.097656). On a fixed natural panel, exploratory PPR used a median closure prefix of 95 rather than 105 for exploratory Serfling/FPC, but still exposed 88.0% of the panel and had no downstream task. Together these results map a restraint--power--cost--utility boundary: the action contract controls a defined persistent decision, while net value depends on evidence margin, audit cost, and budget regime.

cs.LG↗

arXiv2026

CARA: Exact Local Repair with Fresh One-Action Certification for Cloud Consolidation

Simulator-based placement pipelines may inspect many repairs but deploy only when several reliability criteria improve together. Reusing search scenes to test the selected action invalidates nominal evidence, while scalarization can trade away the weakest criterion. We introduce Certificate-Aligned Recomposition (CARA), an incumbent-anchored pipeline that separates adaptive proposal generation from a one-use deployment decision. In a bounded two-host neighborhood, a packing-specific admissible bound recovers the exact top-$P$ distinct actions under a mixed paired-binary/fixed-bet certificate order. Held-out views then select one action and commit its betting plans before fresh paired Certification opens. Under the stated sign-independence and conditional-moment assumptions, the probability of falsely declaring four-way improvement is at most $.05$, irrespective of the size or complexity of upstream search. In a prospectively frozen study over 128 independent synthetic environments and four repeated contexts, the complete fail-closed terminal policy improved a development-selected same-host-count incumbent on all five Evaluation endpoints in every environment. $J$ fell by 3.57 percentage points and the continuous burdens by 21--28%. A matched ordering sensitivity produced absolute mean gaps below $5.3\times10^{-4}$ and does not establish order superiority. CARA thus couples exact auditable proposal recovery to selection-robust fresh certification for last-mile local repair.

cs.DC↗

arXiv2026

Training Data Selection with Gradient Orthogonality for Efficient Domain Adaptation

Fine-tuning large language models (LLMs) for specialized domains often necessitates a trade-off between acquiring domain expertise and retaining general reasoning capabilities, a phenomenon known as catastrophic forgetting. Existing remedies face a dichotomy: gradient surgery methods offer geometric safety but incur prohibitive computational costs via online projections, while efficient data selection approaches reduce overhead but remain blind to conflict-inducing gradient directions. In this paper, we propose Orthogonal Gradient Selection (OGS), a data-centric method that harmonizes domain performance, general capability retention, and training efficiency. OGS shifts the geometric insights of gradient projection from the optimizer to the data selection stage by treating data selection as a constrained decision-making process. By leveraging a lightweight Navigator model and reinforcement learning techniques, OGS dynamically identifies training samples whose gradients are orthogonal to a general-knowledge anchor. This approach ensures naturally safe updates for target models without modifying the optimizer or incurring runtime projection costs. Experiments across medical, legal, and financial domains demonstrate that OGS achieves excellent results, significantly improving domain performance and training efficiency while maintaining or even enhancing performance on general tasks such as GSM8K.

cs.LG↗

arXiv2026

Nearly Optimal Bayesian Inference for Structural Missingness

Structural missingness breaks 'just impute and train': values can be undefined by causal or logical constraints, and the mask may depend on observed variables, unobserved variables (MNAR), and other missingness indicators. It simultaneously brings (i) a catch-22 situation with causal loop, prediction needs the missing features, yet inferring them depends on the missingness mechanism, (ii) under MNAR, the unseen are different, the missing part can come from a shifted distribution, and (iii) plug-in imputation, a single fill-in can lock in uncertainty and yield overconfident, biased decisions. In the Bayesian view, prediction via the posterior predictive distribution integrates over the full model posterior uncertainty, rather than relying on a single point estimate. This framework decouples (i) learning an in-model missing-value posterior from (ii) label prediction by optimizing the predictive posterior distribution, enabling posterior integration. This decoupling yields an in-model almost-free-lunch: once the posterior is learned, prediction is plug-and-play while preserving uncertainty propagation. It achieves SOTA on 43 classification and 15 imputation benchmarks, with finite-sample near Bayes-optimality guarantees under our SCM prior.

cs.LG↗

arXiv2025

Adaptive Data Selection for Multi-Layer Perceptron Training: A Sub-linear Value-Driven Method

Data selection is one of the fundamental problems in neural network training, particularly for multi-layer perceptrons (MLPs) where identifying the most valuable training samples from massive, multi-source, and heterogeneous data sources under budget constraints poses significant challenges. Existing data selection methods, including coreset construction, data Shapley values, and influence functions, suffer from critical limitations: they oversimplify nonlinear transformations, ignore informative intermediate representations in hidden layers, or fail to scale to larger MLPs due to high computational complexity. In response, we propose DVC (Data Value Contribution), a novel budget-aware method for evaluating and selecting data for MLP training that accounts for the dynamic evolution of network parameters during training. The DVC method decomposes data contribution into Layer Value Contribution (LVC) and Global Value Contribution (GVC), employing six carefully designed metrics and corresponding efficient algorithms to capture data characteristics across three dimensions--quality, relevance, and distributional diversity--at different granularities. DVC integrates these assessments with an Upper Confidence Bound (UCB) algorithm for adaptive source selection that balances exploration and exploitation. Extensive experiments across six datasets and eight baselines demonstrate that our method consistently outperforms existing approaches under various budget constraints, achieving superior accuracy and F1 scores. Our approach represents the first systematic treatment of hierarchical data evaluation for neural networks, providing both theoretical guarantees and practical advantages for large-scale machine learning systems.

cs.LG↗

arXiv2025

The Role of r-Modes in Pulsar Spin-down, Pulsar Timing, and Gravitational Waves

We investigate the role of r-mode oscillations in pulsar spin-down and their implications for gravitational wave emission and pulsar timing analysis. Using a non-linear differential framework that includes r-mode contributions, we derive time-dependent solutions for rotational frequency and period evolution. These expressions are validated using observational data from the Crab pulsar with high precision. By analytically fitting braking indices and spin-down coefficients, we link measurable pulsar properties to gravitational wave signatures. Furthermore, we present closed-form expressions for neutron star compactness and tidal deformability using Lambert W and Lambert-Tsallis functions, enabling model-independent inferences from r-mode gravitational wave frequencies. Our results show that incorporating r-modes significantly improves the accuracy of spin-down models and continuous wave detectability, particularly through the inclusion of high-order frequency terms. This framework supports the modeling of timing residuals, glitch quantification, and gravitational wave constraints. Our findings have direct relevance for data analysis in ongoing and future gravitational wave observatories.

astro-ph.HE↗

arXiv2025

YH-MINER: Multimodal Intelligent System for Natural Ecological Reef Metric Extraction

Coral reefs, crucial for sustaining marine biodiversity and ecological processes (e.g., nutrient cycling, habitat provision), face escalating threats, underscoring the need for efficient monitoring. Coral reef ecological monitoring faces dual challenges of low efficiency in manual analysis and insufficient segmentation accuracy in complex underwater scenarios. This study develops the YH-MINER system, establishing an intelligent framework centered on the Multimodal Large Model (MLLM) for "object detection-semantic segmentation-prior input". The system uses the object detection module (mAP@0.5=0.78) to generate spatial prior boxes for coral instances, driving the segment module to complete pixel-level segmentation in low-light and densely occluded scenarios. The segmentation masks and finetuned classification instructions are fed into the Qwen2-VL-based multimodal model as prior inputs, achieving a genus-level classification accuracy of 88% and simultaneously extracting core ecological metrics. Meanwhile, the system retains the scalability of the multimodal model through standardized interfaces, laying a foundation for future integration into multimodal agent-based underwater robots and supporting the full-process automation of "image acquisition-prior generation-real-time analysis".

cs.CV↗

arXiv2025

Unsupervised Multi-modal Feature Alignment for Time Series Representation Learning

In recent times, the field of unsupervised representation learning (URL) for time series data has garnered significant interest due to its remarkable adaptability across diverse downstream applications. Unsupervised learning goals differ from downstream tasks, making it tricky to ensure downstream task utility by focusing only on temporal feature characterization. Researchers have proposed multiple transformations to extract discriminative patterns implied in informative time series, trying to fill the gap. Despite the introduction of a variety of feature engineering techniques, e.g. spectral domain, wavelet transformed features, features in image form and symbolic features etc. the utilization of intricate feature fusion methods and dependence on heterogeneous features during inference hampers the scalability of the solutions. To address this, our study introduces an innovative approach that focuses on aligning and binding time series representations encoded from different modalities, inspired by spectral graph theory, thereby guiding the neural encoder to uncover latent pattern associations among these multi-modal features. In contrast to conventional methods that fuse features from multiple modalities, our proposed approach simplifies the neural architecture by retaining a single time series encoder, consequently leading to preserved scalability. We further demonstrate and prove mechanisms for the encoder to maintain better inductive bias. In our experimental evaluation, we validated the proposed method on a diverse set of time series datasets from various domains. Our approach outperforms existing state-of-the-art URL methods across diverse downstream tasks.

cs.LG↗

arXiv2021

Salience-Aware Event Chain Modeling for Narrative Understanding

Storytelling, whether via fables, news reports, documentaries, or memoirs, can be thought of as the communication of interesting and related events that, taken together, form a concrete process. It is desirable to extract the event chains that represent such processes. However, this extraction remains a challenging problem. We posit that this is due to the nature of the texts from which chains are discovered. Natural language text interleaves a narrative of concrete, salient events with background information, contextualization, opinion, and other elements that are important for a variety of necessary discourse and pragmatics acts but are not part of the principal chain of events being communicated. We introduce methods for extracting this principal chain from natural language text, by filtering away non-salient events and supportive sentences. We demonstrate the effectiveness of our methods at isolating critical event chains by comparing their effect on downstream tasks. We show that by pre-training large language models on our extracted chains, we obtain improvements in two tasks that benefit from a clear understanding of event chains: narrative prediction and event-based temporal question answering. The demonstrated improvements and ablative studies confirm that our extraction method isolates critical event chains.

cs.CL↗

arXiv2021

Social Media Study of Public Opinions on Potential COVID-19 Vaccines: Informing Dissent, Disparities, and Dissemination

The current development of vaccines for SARS-CoV-2 is unprecedented. Little is known, however, about the nuanced public opinions on the vaccines on social media. We adopt a human-guided machine learning framework using more than six million tweets from almost two million unique Twitter users to capture public opinions on the potential vaccines for SARS-CoV-2, classifying them into three groups: pro-vaccine, vaccine-hesitant, and anti-vaccine. We aggregate opinions at the state and country levels, and find that the major changes in the percentages of different opinion groups roughly correspond to the major pandemic-related events. Interestingly, the percentage of the pro-vaccine group is lower in the Southeast part of the United States. Using multinomial logistic regression, we compare demographics, social capital, income, religious status, political affiliations, geo-locations, sentiment of personal pandemic experience and non-pandemic experience, and county-level pandemic severity perception of these three groups to investigate the scope and causes of public opinions on vaccines. We find that socioeconomically disadvantaged groups are more likely to hold polarized opinions on potential COVID-19 vaccines. People who have the worst personal pandemic experience are more likely to hold the anti-vaccine opinion. Next, by conducting counterfactual analyses, we find that the U.S. public is most concerned about the safety, effectiveness, and political issues regarding potential vaccines for COVID-19, and improving personal pandemic experience increases the vaccine acceptance level. We believe this is the first large-scale social media-based study to analyze public opinions on potential COVID-19 vaccines that can inform more effective vaccine distribution policies and strategies.

cs.SI↗

arXiv2020

Monitoring Depression Trend on Twitter during the COVID-19 Pandemic

The COVID-19 pandemic has severely affected people's daily lives and caused tremendous economic loss worldwide. However, its influence on people's mental health conditions has not received as much attention. To study this subject, we choose social media as our main data resource and create by far the largest English Twitter depression dataset containing 2,575 distinct identified depression users with their past tweets. To examine the effect of depression on people's Twitter language, we train three transformer-based depression classification models on the dataset, evaluate their performance with progressively increased training sizes, and compare the model's "tweet chunk"-level and user-level performances. Furthermore, inspired by psychological studies, we create a fusion classifier that combines deep learning model scores with psychological text features and users' demographic information and investigate these features' relations to depression signals. Finally, we demonstrate our model's capability of monitoring both group-level and population-level depression trends by presenting two of its applications during the COVID-19 pandemic. We hope this study can raise awareness among researchers and the general public of COVID-19's impact on people's mental health.

cs.SI↗

arXiv2017

Tactics and Tallies: A Study of the 2016 U.S. Presidential Campaign Using Twitter 'Likes'

We propose a framework to measure, evaluate, and rank campaign effectiveness in the ongoing 2016 U.S. presidential election. Using Twitter data collected from Sept. 2015 to Jan. 2016, we first uncover the tweeting tactics of the candidates and second, using negative binomial regression and exploiting the variations in 'likes,' we evaluate the effectiveness of these tactics. Thirdly, we rank the candidates' campaign tactics by calculating the conditional expectation of their generated 'likes.' We show that while Ted Cruz and Marco Rubio put much weight on President Obama, this tactic is not being well received by their supporters. We demonstrate that Hillary Clinton's tactic of linking herself to President Obama resonates well with her supporters but the same is not true for Bernie Sanders. In addition, we show that Donald Trump is a major topic for all the other candidates and that the women issue is equally emphasized in Sanders' campaign as in Clinton's. Finally, we suggest two ways that politicians can use the feedback mechanism in social media to improve their campaign: (1) use feedback from social media to improve campaign tactics within social media; (2) prototype policies and test the public response from the social media.

cs.SI↗

arXiv2017

When Follow is Just One Click Away: Understanding Twitter Follow Behavior in the 2016 U.S. Presidential Election

Motivated by the two paradoxical facts that the marginal cost of following one extra candidate is close to zero and that the majority of Twitter users choose to follow only one or two candidates, we study the Twitter follow behaviors observed in the 2016 U.S. presidential election. Specifically, we complete the following tasks: (1) analyze Twitter follow patterns of the presidential election on Twitter, (2) use negative binomial regression to study the effects of gender and occupation on the number of candidates that one follows, and (3) use multinomial logistic regression to investigate the effects of gender, occupation and celebrities on the choice of candidates to follow.

cs.SI↗

arXiv2016

Tactics and Tallies: Inferring Voter Preferences in the 2016 U.S. Presidential Primaries Using Sparse Learning

In this paper, we propose a web-centered framework to infer voter preferences for the 2016 U.S. presidential primaries. Using Twitter data collected from Sept. 2015 to March 2016, we first uncover the tweeting tactics of the candidates and then exploit the variations in the number of 'likes' to infer voters' preference. With sparse learning, we are able to reveal neutral topics as well as positive and negative ones. Methodologically, we are able to achieve a higher predictive power with sparse learning. Substantively, we show that for Hillary Clinton the (only) positive issue area is women's rights. We demonstrate that Hillary Clinton's tactic of linking herself to President Obama resonates well with her supporters but the same is not true for Bernie Sanders. In addition, we show that Donald Trump is a major topic for all the other candidates, and that the women's rights issue is equally emphasized in Sanders' campaign as in Clinton's.

cs.SI↗

arXiv2016

Gender Politics in the 2016 U.S. Presidential Election: A Computer Vision Approach

Gender is playing an important role in the 2016 U.S. presidential election, especially with Hillary Clinton becoming the first female presidential nominee and Donald Trump being frequently accused of sexism. In this paper, we introduce computer vision to the study of gender politics and present an image-driven method that can measure the effects of gender in an accurate and timely manner. We first collect all the profile images of the candidates' Twitter followers. Then we train a convolutional neural network using images that contain gender labels. Lastly, we classify all the follower and unfollower images. Through two case studies, one on the `woman card' controversy and one on Sanders followers, we demonstrate how gender is informing the 2016 presidential election. Our framework of analysis can be readily generalized to other case studies and elections.

cs.SI↗

arXiv2016

Will Sanders Supporters Jump Ship for Trump? Fine-grained Analysis of Twitter Followers

In this paper, we study the likelihood of Bernie Sanders supporters voting for Donald Trump instead of Hillary Clinton. Building from a unique time-series dataset of the three candidates' Twitter followers, which we make public here, we first study the proportion of Sanders followers who simultaneously follow Trump (but not Clinton) and how this evolves over time. Then we train a convolutional neural network to classify the gender of Sanders followers, and study whether men are more likely to jump ship for Trump than women. Our study shows that between March and May an increasing proportion of Sanders followers are following Trump (but not Clinton). The proportion of Sanders followers who follow Clinton but not Trump has actually decreased. Equally important, our study suggests that the jumping ship behavior will be affected by gender and that men are more likely to switch to Trump than women.

cs.SI↗

arXiv2016

Pricing the Woman Card: Gender Politics between Hillary Clinton and Donald Trump

In this paper, we propose a data-driven method to measure the impact of the 'woman card' exchange between Hillary Clinton and Donald Trump. Building from a unique dataset of the two candidates' Twitter followers, we first examine the transition dynamics of the two candidates' Twitter followers one week before the exchange and one week after. Then we train a convolutional neural network to classify the gender of the followers and unfollowers, and study how women in particular are reacting to the 'woman card' exchange. Our study suggests that the 'woman card' comment has made women more likely to follow Hillary Clinton, less likely to unfollow her and that it has apparently not affected the gender composition of Trump followers.

cs.SI↗

arXiv2016

Voting with Feet: Who are Leaving Hillary Clinton and Donald Trump?

From a crowded field with 17 candidates, Hillary Clinton and Donald Trump have emerged as the two front-runners in the 2016 U.S. presidential campaign. The two candidates each boast more than 5 million followers on Twitter, and at the same time both have witnessed hundreds of thousands of people leave their camps. In this paper we attempt to characterize individuals who have left Hillary Clinton and Donald Trump between September 2015 and March 2016. Our study focuses on three dimensions of social demographics: social capital, gender, and age. Within each camp, we compare the characteristics of the current followers with former followers, i.e., individuals who have left since September 2015. We use the number of followers to measure social capital, and profile images to infer gender and age. For classifying gender, we train a convolutional neural network (CNN). For age, we use the Face++ API. Our study shows that for both candidates followers with more social capital are more likely to leave (or switch camps). For both candidates females make up a larger presence among unfollowers than among current followers. Somewhat surprisingly, the effect is particularly pronounced for Clinton. Lastly, middle-aged individuals are more likely to leave Trump, and the young are more likely to leave Hillary Clinton.

cs.SI↗