Search

Diffusion models have demonstrated effectiveness in generating natural images and have been extended to generate diverse data types, including graphs. This new generation of diffusion-based graph generative models has demonstrated significant performance improvements over methods that rely on variational autoencoders or generative adversarial networks. It's important to recognize, however, that most of these models employ Gaussian or categorical diffusion processes, which can struggle with sparse and long-tailed data distributions. In our work, we introduce Graph Beta Diffusion (GBD), a diffusion-based generative model particularly adept at capturing diverse graph structures. GBD utilizes a beta diffusion process, tailored for the sparse and range-bounded characteristics of graph adjacency matrices. Furthermore, we have developed a modulation technique that enhances the realism of the generated graphs by stabilizing the generation of critical graph structures, while preserving flexibility elsewhere. The outstanding performance of GBD across three general graph benchmarks and two biochemical graph benchmarks highlights its capability to effectively capture the complexities of real-world graph data. The code will be made available at https://github.com/YH-UtMSB/Graph_Beta_Diffusion

Single-shot ultrafast optical imaging plays a very important role in the detection of transient scenes, especially in capturing irreversible or stochastic dynamic scenes. To break the limit of time response speed of electronic devices, such as charge-coupled device (CCD) or complementary metal-oxide-semiconductor (CMOS) detectors, ultrafast optical imaging techniques usually convert the time information of a transient scene into the wavelength, angle, space or spatial frequency of the illumination light in previous studies. In this work, we propose a novel polarization-resolved ultrafast mapping photography (PUMP) technique by converting the time information into the polarization. Here, the spatiotemporal information of a dynamic scene is loaded into a rotationally polarized illumination laser pulse, and a polarization filtering in imaging detection and a deconvolution algorithm in image reconstruction are used to extract the original dynamic scene. In our PUMP system, the temporal resolution is 850 fs, the spatial resolution is 28.5 lp/mm at 700 micrometer by 700 micrometer field of view, and the number of frames is 16. By using PUMP, a spatiotemporal dynamics of femtosecond laser ablation in an indium tin oxide film on glass substrate is successfully captured. PUMP provides a new solution for measuring the transient scenes in a snapshot, which will bring a very wide range of applications in the field of ultrafast science.

Graph neural networks (GNNs), which propagate the node features through the edges and learn how to transform the aggregated features under label supervision, have achieved great success in supervised feature extraction for both node-level and graph-level classification tasks. However, GNNs typically treat the graph structure as given and ignore how the edges are formed. This paper introduces a graph generative process to model how the observed edges are generated by aggregating the node interactions over a set of overlapping node communities, each of which contributes to the edges via a logical OR mechanism. Based on this generative model, we partition each edge into the summation of multiple community-specific weighted edges and use them to define community-specific GNNs. A variational inference framework is proposed to jointly learn a GNN-based inference network that partitions the edges into different communities, these community-specific GNNs, and a GNN-based predictor that combines community-specific GNNs for the end classification task. Extensive evaluations on real-world graph datasets have verified the effectiveness of the proposed method in learning discriminative representations for both node-level and graph-level classification tasks., Comment: 10 pages, 5 figures, 14 pages of appendix, accepted to NeurIPS 2022

The rotational Doppler effect associated with light's orbital angular momentum (OAM) has been found as a powerful tool to detect rotating bodies. However, this method was only demonstrated experimentally on the laboratory scale under well controlled conditions so far. And its real potential lies at the practical applications in the field of remote sensing. We have established a 120-meter long free-space link between the rooftops of two buildings and show that both the rotation speed and the rotational symmetry of objects can be identified from the detected rotational Doppler frequency shift signal at photon count level. Effects of possible slight misalignments and atmospheric turbulences are quantitatively analyzed in terms of mode power spreading to the adjacent modes as well as the transfer of rotational frequency shifts. Moreover, our results demonstrate that with the preknowledge of the object's rotational symmetry one may always deduce the rotation speed no matter how strong the coupling to neighboring modes is. Without any information of the rotating object, the deduction of the object's symmetry and rotational speed may still be obtained as long as the mode spreading efficiency does not exceed 50 %. Our work supports the feasibility of a practical sensor to remotely detect both the speed and symmetry of rotating bodies.