187037 research outputs found
Sort by
Bridging the Skeleton-Text Modality Gap: Diffusion-Powered Modality Alignment for Zero-shot Skeleton-based Action Recognition
No full textIn zero-shot skeleton-based action recognition (ZSAR), aligning skeleton features with the text features of action labels is essential for accurately predicting unseen actions. ZSAR faces a fundamental challenge in bridging the modality gap between the two-kind features, which severely limits generalization to unseen actions. Previous methods focus on direct alignment between skeleton and text latent spaces, but the modality gaps between these spaces hinder robust generalization learning. Motivated by the success of diffusion models in multi-modal alignment (e.g., text-to-image, text-to-video), we firstly present a diffusion-based skeleton-text alignment framework for ZSAR. Our approach, Triplet Diffusion for Skeleton-Text Matching (TDSM), focuses on cross-alignment power of diffusion models rather than their generative capability. Specifically, TDSM aligns skeleton features with text prompts by incorporating text features into the reverse diffusion process, where skeleton features are denoised under text guidance, forming a unified skeleton-text latent space for robust matching. To enhance discriminative power, we introduce a triplet diffusion (TD) loss that encourages our TDSM to correct skeleton-text matches while pushing them apart for different action classes. Our TDSM significantly outperforms very recent state-of-the-art methods with significantly large margins of 2.36%-point to 13.05%-point, demonstrating superior accuracy and scalability in zero-shot settings through effective skeleton-text matching
A probabilistic AI-based pedestrian volume estimation model for street-level urban management
No full textAs analyzing and controlling human movements is a key task in urban planning and management process, decision-makers have developed the computer-aided methodologies for pedestrian volume estimation. However, the existing pedestrian volume estimation studies have faced several constraints, such as the high cost of manual counting. This study aims to develop a street-level pedestrian volume estimation model by utilizing a computeraided approach to evaluate data acquired through closed-circuit television (CCTV). We employ spatial characteristics and pedestrian volume data derived from sensed points to estimate pedestrian volumes on unsensed streets. The Bayesian Neural Network (BNN)-based model exhibited the best performance for the pedestrian volume estimation, with a mean hourly error of 21.305 individuals for 10-fold cross-validation and total 77.394 individuals for test estimation on four streets representing diverse pedestrian volume, surpassing the performance of regression and deep neural network (DNN)-based models. The BNN-based model also exhibits considerable performance in terms of estimating the hourly pedestrian volumes of entire streets within the study site (i.e., Bukchon Hanok Village in Seoul), and we examine the potential of the stochastic model to assess the uncertainty of a probabilistic artificial intelligence (AI) model. Additionally, a scenario-based test that assumes two specific road control situations presents a numeric and visual aid for predicting the aftermath of an urban management policy. The methodology proposed in this study can contribute to the exploitation of computeraided techniques and trustworthy AI for supporting decision-making processes, thereby providing urban planners with more quantitative evidence.
Impact of diffusion mechanisms on persistence and spreading
No full textWe examine a generalized KPP equation with a q-diffusion, which is a framework that unifies various standard linear diffusion regimes: Fickian diffusion (q = 0), Stratonovich diffusion (q = 1/2), Fokker-Planck diffusion (q = 1), and nonstandard diffusion regimes for general q is an element of & Ropf;. Using both analytical methods and numerical simulations, we explore how the ability of persistence (measured by some principal eigenvalue) and the asymptotic spreading speed depend on the parameter q and on the phase shift between the growth rate r(x) and the diffusion coefficient D(x). Our results demonstrate that persistence and spreading properties generally depend on q. For example, appropriate configurations of r(x) and D(x) can be constructed such that q-diffusion either enhances or diminishes the ability of persistence and the spreading speed with respect to the traditional Fickian diffusion. We show that the spatial arrangement of r(x) with respect to D(x) has markedly different effects depending on whether q > 0, q = 0, or q < 0. The case where r is constant is an exception: persistence becomes independent of q, while the spreading speed displays a symmetry around q = 1/2. This work underscores the importance of carefully selecting diffusion models in ecological and epidemiological contexts, highlighting their potential implications for persistence, spreading, and control strategies.
Laterally Confined Monolayer WS2 Nanodots for Enhanced Excitonic Interaction
No full textMonolayer transition-metal dichalcogenides (TMDCs) host tightly bound excitons with unique valley pseudospin properties, establishing them as an emerging material platform for nanophotonics and quantum technologies. Exciton-exciton interactions modify light-matter coupling and significantly affect the formation of exciton complexes. Here, we employ a top-down nanofabrication technique to manipulate interexcitonic interactions in WS2 monolayers through lateral confinement. By restricting the motion of excitons in confined two-dimensional (2D) spaces, interexcitonic interactions are significantly modified, resulting in strong biexcitonic emission in nanodots smaller than 100 nm that is obscured in pristine monolayers. Moreover, we demonstrate selective optical excitation of valley pseudospins for excitonic quasiparticles in confined monolayers. Our work highlights the role of spatial confinement in excitonic behavior in 2D systems and provides new insights into the development of future photonic and valleytronic devices with low-dimensional platforms.
A federated learning framework for arbitrary spatio-temporal graph neural networks
No full textThe proliferation of mobile and Internet of Things (IoT) devices has resulted in a surge of time-series sensor data, posing significant challenges for centralized data collection and processing. This challenge has driven the adoption of edge computing, which offloads data processing to mid-level servers located at the edge of the Internet, thereby reducing computation and bandwidth demands. Federated learning has emerged as a promising method for training models in edge-computing environments. Recently, spatio-temporal graph neural networks (STGNNs) have shown impressive performance in time-series prediction, yet their application in edge computing is limited by the complexity of adapting them to distributed environments. To address this gap, we propose FedSTGNN (Federated Spatio-Temporal Graph Neural Network), a universal framework that converts existing centralized STGNN models into a federated learning version. We formulate the common STGNN training process using matrix operations, employ graph-based imputation methods to handle missing sensor values at edge servers, and facilitate the transition from centralized to federated STGNNs. Our comprehensive evaluations demonstrate that FedSTGNN not only preserves the prediction accuracy of the original STGNN models but is also significantly more network-efficient than the competing model. Furthermore, the framework proves its robustness in challenging real-world scenarios, including sparse graphs, long-term forecasting, and dynamic server participation. Our work presents a practical, robust, and universal solution for deploying STGNNs into various edge computing applications.
Bi-Modal Learning for Networked Time Series
No full textUnderstanding human mobility patterns is a complex challenge that requires modeling both node-oriented time series (e.g., population) and edge-oriented time series (e.g., population flows) within graph topologies across time. While previous methods have focused on either node-oriented time series or interactions, the synergistic integration of these two modalities has proven difficult to achieve. In this paper, we propose BINTS (BI-modal learning for Networked Time Series), a novel bi-modal learning framework that employs soft contrastive learning along the temporal axis. BINTS captures modality similarities and temporal patterns by simultaneously learning from evolving node-oriented time series and interactions, solving the limitations of single-modality approaches. To evaluate our method, we curate comprehensive multi-modal human mobility datasets spanning diverse locations and times. Our experimental results demonstrate that BINTS significantly outperforms existing forecasting models by capturing synergies across different data modalities. Overall, we establish BINTS as a powerful technique for holistically understanding and forecasting complex mobility dynamics
