Search

While Large Vision-Language Models (LVLMs) have rapidly advanced in recent years, the prevalent issue known as the `hallucination' problem has emerged as a significant bottleneck, hindering their real-world deployments. Existing methods mitigate this issue mainly from two perspectives: One approach leverages extra knowledge like robust instruction tuning LVLMs with curated datasets or employing auxiliary analysis networks, which inevitable incur additional costs. Another approach, known as contrastive decoding, induces hallucinations by manually disturbing the vision or instruction raw inputs and mitigates them by contrasting the outputs of the disturbed and original LVLMs. However, these approaches rely on empirical holistic input disturbances and double the inference cost. To avoid these issues, we propose a simple yet effective method named Self-Introspective Decoding (SID). Our empirical investigation reveals that pretrained LVLMs can introspectively assess the importance of vision tokens based on preceding vision and text (both instruction and generated) tokens. We develop the Context and Text-aware Token Selection (CT2S) strategy, which preserves only unimportant vision tokens after early layers of LVLMs to adaptively amplify text-informed hallucination during the auto-regressive decoding. This approach ensures that multimodal knowledge absorbed in the early layers induces multimodal contextual rather than aimless hallucinations. Subsequently, the original token logits subtract the amplified vision-and-text association hallucinations, guiding LVLMs decoding faithfully. Extensive experiments illustrate SID generates less-hallucination and higher-quality texts across various metrics, without extra knowledge and much additional computation burdens.

Multivariate time series prediction is widely used in daily life, which poses significant challenges due to the complex correlations that exist at multi-grained levels. Unfortunately, the majority of current time series prediction models fail to simultaneously learn the correlations of multivariate time series at multi-grained levels, resulting in suboptimal performance. To address this, we propose a Multi-Grained Correlations-based Prediction (MGCP) Network, which simultaneously considers the correlations at three granularity levels to enhance prediction performance. Specifically, MGCP utilizes Adaptive Fourier Neural Operators and Graph Convolutional Networks to learn the global spatiotemporal correlations and inter-series correlations, enabling the extraction of potential features from multivariate time series at fine-grained and medium-grained levels. Additionally, MGCP employs adversarial training with an attention mechanism-based predictor and conditional discriminator to optimize prediction results at coarse-grained level, ensuring high fidelity between the generated forecast results and the actual data distribution. Finally, we compare MGCP with several state-of-the-art time series prediction algorithms on real-world benchmark datasets, and our results demonstrate the generality and effectiveness of the proposed model.

Detecting damage in critical structures using monitored data is a fundamental task of structural health monitoring, which is extremely important for maintaining structures' safety and life-cycle management. Based on statistical pattern recognition paradigm, damage detection can be conducted by assessing changes in the distribution of properly extracted damage-sensitive features (DSFs). This can be naturally formulated as a distributional change-point detection problem. A good change-point detector for damage detection should be scalable to large DSF datasets, applicable to different types of changes, and capable of controlling for false-positive indications. This study proposes a new distributional change-point detection method for damage detection to address these challenges. We embed the elements of a DSF distributional sequence into the Wasserstein space and construct a moving sum (MOSUM) multiple change-point detector based on Fr\'echet statistics and establish theoretical properties. Extensive simulation studies demonstrate the superiority of our proposed approach against other competitors to address the aforementioned practical requirements. We apply our method to the cable-tension measurements monitored from a long-span cable-stayed bridge for cable damage detection. We conduct a comprehensive change-point analysis for the extracted DSF data, and reveal interesting patterns from the detected changes, which provides valuable insights into cable system damage.

In this paper, by slightly generalizing the notion of 'proposition' in 'Propositional Logic and Modal Logic - A Connection via Relational Semantics' by Shengyang Zhong, we propose a relational semantics of propositional language with bottom, conjunction and imlication, which unifies the relational semantics of intuitionistic logic, Visser's basic propositional logic and orthologic with strict implication. We study the semantic and syntactic consequence relations and prove the soundness and completeness theorems for eight propositional logics.

Diagnosing the changes of structural behaviors using monitoring data is an important objective of structural health monitoring (SHM). The changes in structural behaviors are usually manifested as the feature changes in monitored structural responses; thus, developing effective methods for automatically detecting such changes is of considerable significance. Existing methods for change detection in SHM are mainly used for scalar or vector data, thus incapable of detecting the changes of the features represented by complex data, e.g., the probability density functions (PDFs). Detecting the abrupt changes occurred in the distributions (represented by PDFs) associated with the damage-sensitive features extracted from SHM data are usually of crucial interest for structural condition assessment; however, the SHM community still lacks effective diagnostic tools for detecting such changes. In this study, a change-point detection method is developed in the functional data-analytic framework for PDF-valued sequence, and it is leveraged to diagnose the distributional information break encountered in structural condition assessment. A major challenge in PDF-valued data modeling or analysis is that the PDFs are special functional data subjecting to nonlinear constraints. To tackle this issue, the PDFs are embedded into the Bayes space, and the associated change-point model is constructed by using the linear structure of the Bayes space; then, a hypothesis testing procedure is presented for distributional change-point detection based on the isomorphic mapping between the Bayes space and a functional linear space. Comprehensive simulation studies are conducted to validate the effectiveness of the proposed method as well as demonstrate its superiority over the competing method. Finally, an application to real SHM data illustrates its practical utility in structural condition assessment.

Distributional data analysis, concerned with statistical analysis and modeling for data objects consisting of random probability density functions (PDFs) in the framework of functional data analysis (FDA), has received considerable interest in recent years. However, many important aspects remain unexplored, such as outlier detection and robustness. Existing functional outlier detection methods are mainly used for ordinary functional data and usually perform poorly when applied to PDFs. To fill this gap, this study focuses on PDF-valued outlier detection, as well as its application in robust distributional regression. Similar to ordinary functional data, detecting the shape outlier masked by the "curve net" formed by the bulk of the PDFs is the major challenge in PDF-outlier detection. To this end, we propose a tree-structured transformation system for feature extraction as well as converting the shape outliers to easily detectable magnitude outliers, relevant outlier detectors are designed for the specific transformed data. A multiple detection strategy is also proposed to account for detection uncertainties and to combine different detectors to form a more reliable detection tool. Moreover, we propose a distributional-regression-based approach for detecting the abnormal associations of PDF-valued two-tuples. As a specific application, the proposed outlier detection methods are applied to robustify a distribution-to-distribution regression method, and we develop a robust estimator for the regression operator by downweighting the detected outliers. The proposed methods are validated and evaluated by extensive simulation studies or real data applications. Relevant comparative studies demonstrate the superiority of the developed outlier detection method with other competitors in distributional outlier detection.

Data loss is a critical problem in structural health monitoring (SHM). Probability distributions play a highly important role in many applications. Improving the quality of distribution estimations made using incomplete samples is highly important. Missing samples can be compensated for by applying conventional missing data restoration methods; however, ensuring that restored samples roughly follow underlying distributions of true missing data remains a challenge. Another strategy involves directly restoring the probability density function (PDF) for a sensor when samples are missing by leveraging distribution information from another sensor with complete data using distribution regression techniques; existing methods include the conventional distribution-to-distribution regression (DDR) and distribution-to-warping function regression (DWR) methods. Due to constraints on PDFs and warping functions, the regression functions of both methods are estimated from the Nadaraya-Watson kernel estimator with relatively low degrees of precision. This article proposes a new indirect distribution-to-distribution regression method in the context of functional data analysis for restoring distributions of missing SHM data. PDFs are transformed to ordinary functions residing in a Hilbert space via the newly proposed log-quantile-density (LQD) transformation; the regression for distributions is realized in the transformed space via a functional regression model constructed based on the theory of Reproducing Kernel Hilbert Space (RKHS), corresponding result is subsequently mapped back to the density space through the inverse LQD transformation. Test results using field monitoring data indicate that the new method significantly outperforms conventional methods in general cases; however, in extrapolation cases, the new method is inferior to the distribution-to-warping function regression method., Comment: This is a manuscript. Readers are suggested to read the formal version published in the journal "Mechanical Systems and Signal Processing", https://doi.org/10.1016/j.ymssp.2018.11.052