A5101708161- Particle physics theoretical and experimental studies
- Quantum Chromodynamics and Particle Interactions
- High-Energy Particle Collisions Research
- Neutrino Physics Research
- Dark Matter and Cosmic Phenomena
- Gamma-ray bursts and supernovae
- Particle Detector Development and Performance
- Advanced Graph Neural Networks
- Astrophysics and Cosmic Phenomena
- Astronomical Observations and Instrumentation
- Complex Network Analysis Techniques
- Atomic and Subatomic Physics Research
- Black Holes and Theoretical Physics
- Graph Theory and Algorithms
- Particle accelerators and beam dynamics
- Solar and Space Plasma Dynamics
- Nuclear physics research studies
- Pulsars and Gravitational Waves Research
- Muon and positron interactions and applications
- Medical Imaging Techniques and Applications
- Quantum, superfluid, helium dynamics
- Advanced NMR Techniques and Applications
- Radiation Detection and Scintillator Technologies
- Astrophysical Phenomena and Observations
- Topological and Geometric Data Analysis
Heilongjiang University of Chinese Medicine
2025
Institute of High Energy Physics
2022-2025
Bridge University
2024-2025
University of Cambridge
2021-2024
University of Cincinnati
2023-2024
New Mexico State University
2024
Colorado State University
2023-2024
Louisiana State University
2024
Carnegie Mellon University
2018-2024
Shenzhen MSU-BIT University
2024
Abstract The SiTian project, designed to utilize 60 telescopes distributed across multiple sites in China, is a next-generation time-domain survey initiative. As pathfinder of the mini (MST) has been proposed and implemented test brain, data pipeline, evaluate feasibility its technology science cases. Mounted at Xinglong Observatory, MST project comprised three 30 cm operated since Nov. 2022. Each telescope possesses large field view, covering 2.29° × 1.53° FOV, mounted g', r', i' filters,...
Vertex based and spectral GSP sampling has been studied recently. The literature recognizes that methods in one domain do not have a counterpart the other domain. This paper shows fact can develop unified graph signal theory with analogous interpretations both domains just like traditional DSP. To achieve it, we introduce shift $M$ acting rather than $A$ acts vertex leads to starts from domain, for example, linear invariant (LSI) filtering is polynomials $P(M)$. We then dual versions each of...
Deep learning, particularly convolutional neural networks (CNNs), have yielded rapid, significant improvements in computer vision and related domains. But conventional deep learning architectures perform poorly when data an underlying graph structure, as social, biological, many other This paper explores 1)how signal processing (GSP) can be used to extend CNN components graphs order improve model performance; 2)how design the architecture based on topology or structure of graph.
The graph convolutional layer is core in the architecture of neural networks (CNNs). In literature, both spectrum domain based and vertex layers have been proposed. This paper analyzes these two types demonstrates that convo-lutional suffers from output inconsistencies when shift matrix has repeated eigenvalues. contrast, consistent inherits local feature extraction property classical CNNs with low computational complexity. Experimental results on different data sets also demonstrate exhibit...
Graph convolutional neural networks (GCNNs) are a powerful extension of deep learning techniques to graph-structured data problems. We empirically evaluate several pooling methods for GCNNs, and combinations those graph with three different architectures: GCN, TAGCN, GraphSAGE. confirm that pooling, especially DiffPool, improves classification accuracy on popular datasets find that, average, TAGCN achieves comparable or better than GCN GraphSAGE, particularly larger sparser structures.
To analyze data supported by arbitrary graphs G, DSP has been extended to Graph Signal Processing (GSP) redefining traditional concepts like shift, filtering, and Fourier transform among others. This paper revisits modulation, convolution, sampling of graph signals as appropriate natural extensions the corresponding concepts. define these for both vertex frequency domains, we associate with generic G its shift A, a spectral M Gs. leads GSP theory that parallels in domain existing domain. The...
Graph Convolutional Neural Networks (graph CNNs) adapt the traditional CNN architecture for use on graphs, replacing convolution layers with graph layers. Although similar in architecture, CNNs are used geometric deep learning whereas conventional grid-based data, such as audio or images, seemingly no direct relationship between two classes of neural networks.This paper shows that under certain conditions can be data a good approximation to CNNs, avoiding need CNNs. We show this by using an...
At the sea level, a phenomenon common with all rocket engines, especially for highly over-expanded nozzle, during ignition and shutdown is that of flow separation as plume fills empties Since will be separated randomly. it generate side loads, i.e. non-axial forces. engines are designed to produce axial thrust power vehicles, not desirable excited by input forcing functions, In past, several engine failures were attributed loads. During development stage, in order design/size components...
Graph signals are indexed by nodes of the graph. Signal Processing (GSP) extends Discrete (DSP) to graph signals.In this paper, we explore shifting, convolution, and modulation for signals. We begin reviewing some basic concepts in GSP. Then, define M, spectral domain shift analogous vertex shift, A. Finally, use A M polynomial filters, P (A) (M), illustrate convolution both domain.
Graph convolutional neural networks (CNNs) use data that is supported on an arbitrary graph rather than a grid. Most current approaches to classification spectral domain based CNNs and do not work for directed graphs. A more recent approach, topology adaptive convolution (TAGCN), uses from signal processing defined in the vertex instead of domain. In this paper, we TAGCN classify different time periods during week New York City taxi graph. We achieve accuracy 88% using single layer. entire...
The ground-state mass excess of the $T_{z}=-2$ drip-line nucleus $^{22}$Al is measured for first time to be $18103(10)$ keV using newly-developed B$\rho$-defined isochronous spectrometry method at cooler storage ring in Lanzhou. new value allowed us determine excitation energies two low-lying $1^+$ states with significantly reduced uncertainties 51 keV. Comparing analogue its mirror $^{22}$F, energy differences $^{22}$Al-$^{22}$F pair are determined $-625(51)$ and $-330(51)$ keV,...
Many Graph Signal Processing (GSP) applications consider product graphs, the of smaller graphs. For example, with time-varying graph data, shift can be (Cartesian) a space and cyclic time shift. Instead treating as single entity applying existing GSP techniques, there are computational experimental advantages to considering its factors.Recently, in [1], we showed that is DSP plus boundary conditions (b.c.) companion model introduced. Under certain conditions, any converted into consisting 1D...
Convolutional neural networks (CNNs) have been very successful with learning on grid-based data such as time series and images. However, traditional CNNs do not perform well irregular-structured defined a graph. Graph convolutional (graph CNNs) define layers using graph signal processing (GSP) concepts. They citation NYC taxi pickup data. Polynomial filter use polynomial of the adjacency matrix A in layer. they shown to fail classification problems when layer produces same output for...
Graph signal processing (GSP) was designed in [1] (see also [2], [3]), as a natural, intuitive extension of traditional discrete (DSP). Concepts such shifting, filtering, graph signal, Fourier transform, and spectral analysis are naturally extended from DSP to GSP. However, other concepts, delta functions, sampling, structure cannot be extended. This illustrates gap between GSP when introducing new concepts GSP: intuitions do not necessarily hold for draw inspiration DSP. The companion model...
In Discrete Signal Processing (DSP), sampling in the time domain chooses samples from which original bandlimited signal can be perfectly reconstructed. Uniform interpreted frequency as replication using a LSI filtering. Sampling Graph (GSP) has been explored either vertex and spectral domains, but not both. Current GSP literature [1] recognizes that is same domain, leading to two different methods (one other domain). Recently [2], we showed one indeed develop an unified theory with...
This paper introduces a $\textit{canonical}$ graph signal model defined by and shift, the $\textit{companion}$ shift. These are canonical because, under standard conditions, we show that any processing (GSP) can be transformed into model. The transform obtains this is $z$-transform ($\textrm{G$z$T}$) introduce. GSP comes closest to discrete (DSP) time models: structure of companion shift decomposes line continuation just like DSP directed with terminal condition reflecting condition. We...
Graph convolutional neural networks (graph CNNs) have shown great promise in graph classification problems with data. Current CNNs use vertex domain layers, defined using a polynomial of the adjacency matrix, A. However, spectral shift, M, comparable or sometimes better performance than their counterparts. The matrix A is usually sparse, while corresponding M dense. In this paper, we explore methods to produce sparser comparing several different ways remove edges and impact on CNN accuracy....