Search

Bayesian inversion based on normal modes analysis is an essential approach to understanding the internal structure of the Earth and other planets. While the traditional Markov Chain Monte Carlo (MCMC) method provides a feasible approach to obtain the posterior probability density distributions of model parameters, it demands a considerable amount of forward calculation and is slow in terms of computational efficiency. In order to overcome the limitations of the MCMC method, a new approach to Bayesian inversion using generative neural networks has been proposed in this study, which can deliver outcomes equivalent in accuracy to those of the MCMC method with less sample learning. The study proposed three types of Bayesian inversion neural network models based on popular generative neural networks, including GAN, flow model, and energy model, and evaluated their accuracy and discreteness of inversion results. By using the generative neural network method based on actual lunar normal mode observation data, this study has provided posterior probability density distribution of one-dimensional density and velocity structure parameters of the Moon, along with the optimal parameter values.The results demonstrate the potential of using generative neural networks in Bayesian inversion and provide a new direction for future research in this field., The 28th IUGG General Assembly (IUGG2023) (Berlin 2023)

The normal modes (i.e. Earth's free oscillations) are long-period low-frequency seismic signals, which are excited by a variety of factors, such as earthquakes, volcanic eruption, landslide, avalanche and so on, are an essential vehicle for global seismic tomography to elucidate large-scale heterogeneities within the deep Earth. Accurate extraction of signals on normal mode spectrum is a prerequisite for the imaging inversion, providing the differences between the observed and synthetic normal mode spectrum. However, the normal mode spectrum has great complexity due to many structural factors within the Earth, so unacceptable false and dismissed selections of the signals always occur, which hinder the development of exploration of the deep Earth’s deep interior based on normal mode data. To address these problems, we build a deep-learning based neural network, named ModeNet, which is capable of precisely and efficient selecting the frequency windows to cover the target normal modal signals on a noisy spectrum, which could outperform the conventional spectrum-FLEXWIN method without relying on comparisons with synthetics. We also define our own method to evaluate the performance of ModeNet on the testing set and obtain a precision as high as ~0.98. Moreover, ModeNet achieves good generalization in processing seismograms of different events with different noise levels, components, and time window data, as well as superconductivity-gravimeter observations. Therefore, ModeNet could be implemented as a valuable tool for the future deep Earth inversion., The 28th IUGG General Assembly (IUGG2023) (Berlin 2023)

MPTCP is a new transport protocol that enables mobile devices to use multiple physical paths simultaneously through several network interfaces, such as WiFi and Cellular. However, wireless path capacities change frequently in the mobile environments, causing challenges for path selection. For example, WiFi associated paths often become poor as devices walk away, since WiFi has intermittent connectivity caused by the short signal coverage and stochastic interference. MPTCP's native decision based on hysteretic TCP-layer estimation will miss the real switching point of wireless quality, which may cumulate packets on the broken path and causes serious packets reinjection. Through analyzing a unique dataset in the wild, we quantitatively study the impact of MAC-layer factors on the aggregated performance of MPTCP. We then propose a decision tree approach for cross-layer path selection that decides which path to carry the incoming packets dynamically according to the prior learned schemes. A prototype of the path selection system named SmartPS, which proactively probes the wireless environments, is realized and deployed in Linux and Android. Evaluation results demonstrate that our SmartPS can efficiently utilize the faster path, with goodput improvements of up to 29%., Comment: ICNP under review