arXiv subjects

Liang Zhan

Publications and source records attributed to Liang Zhan.

At least 55 records · Page 3Linked to original sources

arXiv2021

Boundary-aware Graph Reasoning for Semantic Segmentation

In this paper, we propose a Boundary-aware Graph Reasoning (BGR) module to learn long-range contextual features for semantic segmentation. Rather than directly construct the graph based on the backbone features, our BGR module explores a reasonable way to combine segmentation erroneous regions with the graph construction scenario. Motivated by the fact that most hard-to-segment pixels broadly distribute on boundary regions, our BGR module uses the boundary score map as prior knowledge to intensify the graph node connections and thereby guide the graph reasoning focus on boundary regions. In addition, we employ an efficient graph convolution implementation to reduce the computational cost, which benefits the integration of our BGR module into current segmentation backbones. Extensive experiments on three challenging segmentation benchmarks demonstrate the effectiveness of our proposed BGR module for semantic segmentation.

cs.CV↗

arXiv2021

PSGR: Pixel-wise Sparse Graph Reasoning for COVID-19 Pneumonia Segmentation in CT Images

Automated and accurate segmentation of the infected regions in computed tomography (CT) images is critical for the prediction of the pathological stage and treatment response of COVID-19. Several deep convolutional neural networks (DCNNs) have been designed for this task, whose performance, however, tends to be suppressed by their limited local receptive fields and insufficient global reasoning ability. In this paper, we propose a pixel-wise sparse graph reasoning (PSGR) module and insert it into a segmentation network to enhance the modeling of long-range dependencies for COVID-19 infected region segmentation in CT images. In the PSGR module, a graph is first constructed by projecting each pixel on a node based on the features produced by the segmentation backbone, and then converted into a sparsely-connected graph by keeping only K strongest connections to each uncertain pixel. The long-range information reasoning is performed on the sparsely-connected graph to generate enhanced features. The advantages of this module are two-fold: (1) the pixel-wise mapping strategy not only avoids imprecise pixel-to-node projections but also preserves the inherent information of each pixel for global reasoning; and (2) the sparsely-connected graph construction results in effective information retrieval and reduction of the noise propagation. The proposed solution has been evaluated against four widely-used segmentation models on three public datasets. The results show that the segmentation model equipped with our PSGR module can effectively segment COVID-19 infected regions in CT images, outperforming all other competing models.

cs.CV↗

arXiv2021

Multiplex Graph Networks for Multimodal Brain Network Analysis

In this paper, we propose MGNet, a simple and effective multiplex graph convolutional network (GCN) model for multimodal brain network analysis. The proposed method integrates tensor representation into the multiplex GCN model to extract the latent structures of a set of multimodal brain networks, which allows an intuitive 'grasping' of the common space for multimodal data. Multimodal representations are then generated with multiplex GCNs to capture specific graph structures. We conduct classification task on two challenging real-world datasets (HIV and Bipolar disorder), and the proposed MGNet demonstrates state-of-the-art performance compared to competitive benchmark methods. Apart from objective evaluations, this study may bear special significance upon network theory to the understanding of human connectome in different modalities. The code is available at https://github.com/ZhaomingKong/MGNets.

cs.CV↗

arXiv2020

CommPOOL: An Interpretable Graph Pooling Framework for Hierarchical Graph Representation Learning

Recent years have witnessed the emergence and flourishing of hierarchical graph pooling neural networks (HGPNNs) which are effective graph representation learning approaches for graph level tasks such as graph classification. However, current HGPNNs do not take full advantage of the graph's intrinsic structures (e.g., community structure). Moreover, the pooling operations in existing HGPNNs are difficult to be interpreted. In this paper, we propose a new interpretable graph pooling framework - CommPOOL, that can capture and preserve the hierarchical community structure of graphs in the graph representation learning process. Specifically, the proposed community pooling mechanism in CommPOOL utilizes an unsupervised approach for capturing the inherent community structure of graphs in an interpretable manner. CommPOOL is a general and flexible framework for hierarchical graph representation learning that can further facilitate various graph-level tasks. Evaluations on five public benchmark datasets and one synthetic dataset demonstrate the superior performance of CommPOOL in graph representation learning for graph classification compared to the state-of-the-art baseline methods, and its effectiveness in capturing and preserving the community structure of graphs.

cs.LG↗

arXiv2020

Deep Representation Learning For Multimodal Brain Networks

Applying network science approaches to investigate the functions and anatomy of the human brain is prevalent in modern medical imaging analysis. Due to the complex network topology, for an individual brain, mining a discriminative network representation from the multimodal brain networks is non-trivial. The recent success of deep learning techniques on graph-structured data suggests a new way to model the non-linear cross-modality relationship. However, current deep brain network methods either ignore the intrinsic graph topology or require a network basis shared within a group. To address these challenges, we propose a novel end-to-end deep graph representation learning (Deep Multimodal Brain Networks - DMBN) to fuse multimodal brain networks. Specifically, we decipher the cross-modality relationship through a graph encoding and decoding process. The higher-order network mappings from brain structural networks to functional networks are learned in the node domain. The learned network representation is a set of node features that are informative to induce brain saliency maps in a supervised manner. We test our framework in both synthetic and real image data. The experimental results show the superiority of the proposed method over some other state-of-the-art deep brain network models.

cs.CV↗

arXiv2020

Adversarial Attack on Hierarchical Graph Pooling Neural Networks

Recent years have witnessed the emergence and development of graph neural networks (GNNs), which have been shown as a powerful approach for graph representation learning in many tasks, such as node classification and graph classification. The research on the robustness of these models has also started to attract attentions in the machine learning field. However, most of the existing work in this area focus on the GNNs for node-level tasks, while little work has been done to study the robustness of the GNNs for the graph classification task. In this paper, we aim to explore the vulnerability of the Hierarchical Graph Pooling (HGP) Neural Networks, which are advanced GNNs that perform very well in the graph classification in terms of prediction accuracy. We propose an adversarial attack framework for this task. Specifically, we design a surrogate model that consists of convolutional and pooling operators to generate adversarial samples to fool the hierarchical GNN-based graph classification models. We set the preserved nodes by the pooling operator as our attack targets, and then we perturb the attack targets slightly to fool the pooling operator in hierarchical GNNs so that they will select the wrong nodes to preserve. We show the adversarial samples generated from multiple datasets by our surrogate model have enough transferability to attack current state-of-art graph classification models. Furthermore, we conduct the robust train on the target models and demonstrate that the retrained graph classification models are able to better defend against the attack from the adversarial samples. To the best of our knowledge, this is the first work on the adversarial attack against hierarchical GNN-based graph classification models.

cs.LG↗

arXiv2020

Improving the Energy Resolution of the Reactor Antineutrino Energy Reconstruction with Positron Direction

The energy resolution is crucial for the reactor neutrino experiments which aims to determine neutrino mass ordering by precise measurement of the reactor antineutrino energy spectrum. A non-negligible effect in the antineutrino energy resolution is the spread of the kinetic energy of the recoiled neutron and the corresponding positron when detecting the antineutrinos via Inverse Beta-Decay (IBD) reaction. The emission direction of the produced positron in IBD reaction can be used to estimate the kinetic energy of neutron and thus the reconstructed antineutrino energy resolution can be improved. To demonstrate the feasibility, a simple positron direction reconstruction method is implemented in a toy liquid scintillator detector like the Taishan Antineutrino Observatory (TAO) with 4500 photoelectron yield per MeV. A 4% to 26% improvement of energy resolution can be achieved for 5 MeV reactor antineutrinos at TAO.

physics.ins-det↗

arXiv2019

Neutrino-based tools for nuclear verification and diplomacy in North Korea

We present neutrino-based options for verifying that the nuclear reactors at North Korea's Yongbyon Nuclear Research Center are no longer operating or that they are operating in an agreed manner, precluding weapons production. Neutrino detectors may be a mutually agreeable complement to traditional verification protocols because they do not require access inside reactor buildings, could be installed collaboratively, and provide persistent and specific observations. At Yongbyon, neutrino detectors could passively verify reactor shutdowns or monitor power levels and plutonium contents, all from outside the reactor buildings. The monitoring options presented here build on recent successes in basic particle physics. Following a dedicated design study, these tools could be deployed in as little as one year at a reasonable cost. In North Korea, cooperative deployment of neutrino detectors could help redirect a limited number of scientists and engineers from military applications to peaceful technical work in an international community. Opportunities for scientific collaboration with South Korea are especially strong. We encourage policymakers to consider collaborative neutrino projects within a broader program of action toward stability and security on the Korean Peninsula.

physics.soc-ph↗

arXiv2018

Investigation of antineutrino spectral anomaly with updated nuclear database

Recently, three successful antineutrino experiments (Daya Bay, Double Chooz, and RENO) measured the neutrino mixing angle theta_13; however, significant discrepancies were found, both in the absolute flux and spectral shape. In this study, the antineutrino spectra were calculated by using the updated nuclear database, and we found that the four isotopes antineutrino spectrum have all contribution to the 5--7 MeV bump with ENDF/B-VII.1 fission yield. The bump can be explained well using the updated library and more important isotopes contribution to the bump were also given. In the last, the fission yield correlation coefficient between the four isotopes were discussed, and found that the correlation coefficients are very large.

nucl-ex↗

arXiv2018

Subspace Network: Deep Multi-Task Censored Regression for Modeling Neurodegenerative Diseases

Over the past decade a wide spectrum of machine learning models have been developed to model the neurodegenerative diseases, associating biomarkers, especially non-intrusive neuroimaging markers, with key clinical scores measuring the cognitive status of patients. Multi-task learning (MTL) has been commonly utilized by these studies to address high dimensionality and small cohort size challenges. However, most existing MTL approaches are based on linear models and suffer from two major limitations: 1) they cannot explicitly consider upper/lower bounds in these clinical scores; 2) they lack the capability to capture complicated non-linear interactions among the variables. In this paper, we propose Subspace Network, an efficient deep modeling approach for non-linear multi-task censored regression. Each layer of the subspace network performs a multi-task censored regression to improve upon the predictions from the last layer via sketching a low-dimensional subspace to perform knowledge transfer among learning tasks. Under mild assumptions, for each layer the parametric subspace can be recovered using only one pass of training data. Empirical results demonstrate that the proposed subspace network quickly picks up the correct parameter subspaces, and outperforms state-of-the-arts in predicting neurodegenerative clinical scores using information in brain imaging.

cs.LG↗

arXiv2018

Sex-by-age differences in the resting-state brain connectivity

Recently we developed a novel method for assessing the hierarchical modularity of functional brain networks - the probability associated community estimation(PACE). The PACE algorithm is unique in that it permits a dual formulation, thus yielding equivalent connectome modular structure regardless of whether considering positive or negative edges. This method was rigorously validated using F1000 and HCP data. We detected novel sex differences in resting-state connectivity that were not previously reported. This current study more thoroughly examined sex differences as a function of age and their clinical correlates, with findings supporting a basal configuration framework. To this end, we found that men and women do not significantly differ in the 22-25 age range. However, these same non-significant differences attained statistical significance in the 26-30 age group, while becoming highly statistically significant in the 31-35 age group. At the most global level, areas of diverging sex difference include parts of the prefrontal cortex and the temporal lobe, amygdala, hippocampus, inferior parietal lobule, posterior cingulate, and precuneus. Further, we identified statistically different self-reported summary scores of inattention, hyperactivity, and anxiety problems between men and women. These self-reports additionally divergently interact with age and the basal configuration between sexes. In sum, our study supports a paradigm change in how we conceptualize the functional connectome, shifting away from simple concepts, and towards thinking globally and probabilistically how the brain exhibits dynamic sex-specific connectivity configuration as a function of age, and the role this sex-by-age configuration at rest might play in mental health frequency and presentation, including symptom patterns in depression.

q-bio.NC↗

arXiv2017

Exploring the Human Connectome Topology in Group Studies

Visually comparing brain networks, or connectomes, is an essential task in the field of neuroscience. Especially relevant to the field of clinical neuroscience, group studies that examine differences between populations or changes over time within a population enable neuroscientists to reason about effective diagnoses and treatments for a range of neuropsychiatric disorders. In this paper, we specifically explore how visual analytics tools can be used to facilitate various clinical neuroscience tasks, in which observation and analysis of meaningful patterns in the connectome can support patient diagnosis and treatment. We conduct a survey of visualization tasks that enable clinical neuroscience activities, and further explore how existing connectome visualization tools support or fail to support these tasks. Based on our investigation of these tasks, we introduce a novel visualization tool, NeuroCave, to support group studies analyses. We discuss how our design decisions (the use of immersive visualization, the use of hierarchical clustering and dimensionality reduction techniques, and the choice of visual encodings) are motivated by these tasks. We evaluate NeuroCave through two use cases that illustrate the utility of interactive connectome visualization in clinical neuroscience contexts. In the first use case, we study sex differences using functional connectomes and discover hidden connectome patterns associated with well-known cognitive differences in spatial and verbal abilities. In the second use case, we show how the utility of visualizing the brain in different topological space coupled with clustering information can reveal the brain's intrinsic structure.

q-bio.NC↗

arXiv2017

The Importance of Being Negative: A serious treatment of non-trivial edges in brain functional connectome

Understanding the modularity of fMRI-derived brain networks or connectomes can inform the study of brain function organization. However, fMRI connectomes additionally involve negative edges, which are not rigorously accounted for by existing approaches to modularity that either ignores or arbitrarily weight these connections. Furthermore, most Q maximization-based modularity algorithms yield variable results with suboptimal reproducibility. Here we present an alternative, reproducible approach that exploits how frequent the BOLD-signal correlation between two nodes is negative. We validated this novel probability-based modularity approach on two independent publicly-available resting-state connectome dataset (the Human Connectome Project and the 1000 Functional Connectomes) and demonstrated that negative correlations alone are sufficient in understanding resting-state modularity. In fact, this approach a) permits a dual formulation, leading to equivalent solutions regardless of whether one considers positive or negative edges; b) is theoretically linked to the Ising model defined on the connectome, thus yielding modularity result that maximizes data likelihood. We additionally were able to detect sex differences in modularity that the most widely utilized methods did not. Results confirmed the superiority of our approach in that: a) correlations with the highest probability of being negative are consistently placed between modules, b) due to the equivalent dual forms, no arbitrary weighting factor is required to balance the influence between negative and positive correlations, unlike existing Q maximization-based modularity approaches. As datasets like HCP become widely available for analysis by the neuroscience community at large, appropriate computational tools to understand the neurobiological information of negative edges in fMRI connectomes are increasingly important.

q-bio.QM↗

arXiv2016

Large-scale Collaborative Imaging Genetics Studies of Risk Genetic Factors for Alzheimer's Disease Across Multiple Institutions

Genome-wide association studies (GWAS) offer new opportunities to identify genetic risk factors for Alzheimer's disease (AD). Recently, collaborative efforts across different institutions emerged that enhance the power of many existing techniques on individual institution data. However, a major barrier to collaborative studies of GWAS is that many institutions need to preserve individual data privacy. To address this challenge, we propose a novel distributed framework, termed Local Query Model (LQM) to detect risk SNPs for AD across multiple research institutions. To accelerate the learning process, we propose a Distributed Enhanced Dual Polytope Projection (D-EDPP) screening rule to identify irrelevant features and remove them from the optimization. To the best of our knowledge, this is the first successful run of the computationally intensive model selection procedure to learn a consistent model across different institutions without compromising their privacy while ranking the SNPs that may collectively affect AD. Empirical studies are conducted on 809 subjects with 5.9 million SNP features which are distributed across three individual institutions. D-EDPP achieved a 66-fold speed-up by effectively identifying irrelevant features.

cs.LG↗

arXiv2016

Potential of Geo-neutrino Measurements at JUNO

The flux of geoneutrinos at any point on the Earth is a function of the abundance and distribution of radioactive elements within our planet. This flux has been successfully detected by the 1-kt KamLAND and 0.3-kt Borexino detectors with these measurements being limited by their low statistics. The planned 20-kt JUNO detector will provide an exciting opportunity to obtain a high statistics measurement, which will provide data to address several questions of geological importance. This paper presents the JUNO detector design concept, the expected geo-neutrino signal and corresponding backgrounds. The precision level of geo-neutrino measurements at JUNO is obtained with the standard least-squares method. The potential of the Th/U ratio and mantle measurements is also discussed.

physics.ins-det↗

arXiv2016

Mass hierarchy sensitivity of medium baseline reactor neutrino experiments with multiple detectors

We report the neutrino mass hierarchy (MH) sensitivity of medium baseline reactor neutrino experiments with multiple detectors. Sensitivity of determining the MH can be significantly improved by adding a near detector and combining both the near and far detectors. The size of the sensitivity improvement is related to accuracy of the individual mass-splitting measurements and requires strict control on the relative energy scale uncertainty of the near and far detectors. We study the impact of both baseline and target mass of the near detector on the combined sensitivity. A figure-of-merit is defined to optimize the baseline and target mass of the near detector and the optimal selections are $\sim$13~km and $\sim$4~kton respectively for a far detector with the 20~kton target mass and 52.5~km baseline. As typical examples of future medium baseline reactor neutrino experiments, the optimal location and target mass of the near detector are selected for JUNO and RENO-50. Finally, we discuss distinct effects of the neutrino spectrum uncertainty for setups of a single detector and double detectors, which indicate that the spectrum uncertainty can be well constrained in the presence of the near detector.

physics.ins-det↗

arXiv2015

Recent Results from Daya Bay

The Daya Bay reactor neutrino experiment observed electron antineutrino disappearance associated with $θ_{13}$ with a significance better than $5σ$ in 2012. The final two of eight antineutrino detectors were installed in the summer of 2012. Including the 404 days of data collected with the full detector configuration resulted in a 3.6 times increase of statistics over the previous result with the 6-AD configuration. With improvements of the systematic uncertainties and better estimation of backgrounds, Daya Bay has measured $\sin^22θ_{13} = 0.084\pm0.005$ and $|Δm^2_{ee}|=2.42^{+0.10}_{-0.11}\times 10^{-3}$~eV$^2$. This is the most precise measurement of $\sin^22θ_{13}$ to date and the most precise measurement of of $|Δm^2_{ee}|$ via electron antineutrino disappearance. Several other analysis results are presented, including an independent measurement of $θ_{13}$ using inverse-beta decays associated with neutron capture on hydrogen, a measurement of reactor antineutrino flux and spectrum, and a search for light sterile neutrino mixing.

hep-ex↗

arXiv2015

JUNO: A Next Generation Reactor Antineutrino Experiment

The mass hierarchy and the CP phase are the main focus of the next generation neutrino oscillation experiments. Jiangmen Underground Neutrino Observatory (JUNO), as a medium baseline reactor antineutrino experiment, can determine the neutrino mass hierarchy independent of the CP phase. The physics potential on the mass hierarchy, and other measurements are reviewed. The preliminary design options for a 20~kton detector with an energy resolution of $3\%/\sqrt{E_{vis}}$ are illustrated. The main technical challenges on the PMT and scintillator are discussed and the corresponding R\&D efforts are presented.

physics.ins-det↗