arXiv subjects

Cheng Peng

Publications and source records attributed to Cheng Peng.

At least 181 records · Page 10Linked to original sources

arXiv2022

Half-Wormholes and Ensemble Averages

We study "half-wormhole-like" saddle point contributions to spectral correlators in a variety of ensemble average models, including various statistical models, generalized 0d SYK models, 1d Brownian SYK models and an extension of it. In statistical ensemble models, where more general distributions of the random variables could be studied in great details, we find the accuracy of the previously proposed approximation for the half-wormholes could be improved when the distribution of the random variables deviate significantly from Gaussian distributions. We propose a modified approximation scheme of the half-wormhole contributions that also work well in these more general theories. In various generalized 0d SYK models we identify new half-wormhole-like saddle point contributions. In the 0d SYK model and 1d Brownian SYK model, apart from the wormhole and half-wormhole saddles, we find new non-trivial saddles in the spectral correlators that would potentially give contributions of the same order as the trivial self-averaging saddles. However after a careful Lefschetz-thimble analysis we show that these non-trivial saddles should not be included. We also clarify the difference between "linked half-wormholes" and "unlinked half-wormholes" in some models.

hep-th↗

arXiv2022

XraySyn: Realistic View Synthesis From a Single Radiograph Through CT Priors

A radiograph visualizes the internal anatomy of a patient through the use of X-ray, which projects 3D information onto a 2D plane. Hence, radiograph analysis naturally requires physicians to relate the prior about 3D human anatomy to 2D radiographs. Synthesizing novel radiographic views in a small range can assist physicians in interpreting anatomy more reliably; however, radiograph view synthesis is heavily ill-posed, lacking in paired data, and lacking in differentiable operations to leverage learning-based approaches. To address these problems, we use Computed Tomography (CT) for radiograph simulation and design a differentiable projection algorithm, which enables us to achieve geometrically consistent transformations between the radiography and CT domains. Our method, XraySyn, can synthesize novel views on real radiographs through a combination of realistic simulation and finetuning on real radiographs. To the best of our knowledge, this is the first work on radiograph view synthesis. We show that by gaining an understanding of radiography in 3D space, our method can be applied to radiograph bone extraction and suppression without groundtruth bone labels.

eess.IV↗

arXiv2022

HyperSegNAS: Bridging One-Shot Neural Architecture Search with 3D Medical Image Segmentation using HyperNet

Semantic segmentation of 3D medical images is a challenging task due to the high variability of the shape and pattern of objects (such as organs or tumors). Given the recent success of deep learning in medical image segmentation, Neural Architecture Search (NAS) has been introduced to find high-performance 3D segmentation network architectures. However, because of the massive computational requirements of 3D data and the discrete optimization nature of architecture search, previous NAS methods require a long search time or necessary continuous relaxation, and commonly lead to sub-optimal network architectures. While one-shot NAS can potentially address these disadvantages, its application in the segmentation domain has not been well studied in the expansive multi-scale multi-path search space. To enable one-shot NAS for medical image segmentation, our method, named HyperSegNAS, introduces a HyperNet to assist super-net training by incorporating architecture topology information. Such a HyperNet can be removed once the super-net is trained and introduces no overhead during architecture search. We show that HyperSegNAS yields better performing and more intuitive architectures compared to the previous state-of-the-art (SOTA) segmentation networks; furthermore, it can quickly and accurately find good architecture candidates under different computing constraints. Our method is evaluated on public datasets from the Medical Segmentation Decathlon (MSD) challenge, and achieves SOTA performances.

eess.IV↗

arXiv2022

Towards performant and reliable undersampled MR reconstruction via diffusion model sampling

Magnetic Resonance (MR) image reconstruction from under-sampled acquisition promises faster scanning time. To this end, current State-of-The-Art (SoTA) approaches leverage deep neural networks and supervised training to learn a recovery model. While these approaches achieve impressive performances, the learned model can be fragile on unseen degradation, e.g. when given a different acceleration factor. These methods are also generally deterministic and provide a single solution to an ill-posed problem; as such, it can be difficult for practitioners to understand the reliability of the reconstruction. We introduce DiffuseRecon, a novel diffusion model-based MR reconstruction method. DiffuseRecon guides the generation process based on the observed signals and a pre-trained diffusion model, and does not require additional training on specific acceleration factors. DiffuseRecon is stochastic in nature and generates results from a distribution of fully-sampled MR images; as such, it allows us to explicitly visualize different potential reconstruction solutions. Lastly, DiffuseRecon proposes an accelerated, coarse-to-fine Monte-Carlo sampling scheme to approximate the most likely reconstruction candidate. The proposed DiffuseRecon achieves SoTA performances reconstructing from raw acquisition signals in fastMRI and SKM-TEA. Code will be open-sourced at www.github.com/cpeng93/DiffuseRecon.

eess.IV↗

arXiv2022

Undersampled MRI Reconstruction with Side Information-Guided Normalisation

Magnetic resonance (MR) images exhibit various contrasts and appearances based on factors such as different acquisition protocols, views, manufacturers, scanning parameters, etc. This generally accessible appearance-related side information affects deep learning-based undersampled magnetic resonance imaging (MRI) reconstruction frameworks, but has been overlooked in the majority of current works. In this paper, we investigate the use of such side information as normalisation parameters in a convolutional neural network (CNN) to improve undersampled MRI reconstruction. Specifically, a Side Information-Guided Normalisation (SIGN) module, containing only few layers, is proposed to efficiently encode the side information and output the normalisation parameters. We examine the effectiveness of such a module on two popular reconstruction architectures, D5C5 and OUCR. The experimental results on both brain and knee images under various acceleration rates demonstrate that the proposed method improves on its corresponding baseline architectures with a significant margin.

eess.IV↗

arXiv2021

RSBNet: One-Shot Neural Architecture Search for A Backbone Network in Remote Sensing Image Recognition

Recently, a massive number of deep learning based approaches have been successfully applied to various remote sensing image (RSI) recognition tasks. However, most existing advances of deep learning methods in the RSI field heavily rely on the features extracted by the manually designed backbone network, which severely hinders the potential of deep learning models due the complexity of RSI and the limitation of prior knowledge. In this paper, we research a new design paradigm for the backbone architecture in RSI recognition tasks, including scene classification, land-cover classification and object detection. A novel one-shot architecture search framework based on weight-sharing strategy and evolutionary algorithm is proposed, called RSBNet, which consists of three stages: Firstly, a supernet constructed in a layer-wise search space is pretrained on a self-assembled large-scale RSI dataset based on an ensemble single-path training strategy. Next, the pre-trained supernet is equipped with different recognition heads through the switchable recognition module and respectively fine-tuned on the target dataset to obtain task-specific supernet. Finally, we search the optimal backbone architecture for different recognition tasks based on the evolutionary algorithm without any network training. Extensive experiments have been conducted on five benchmark datasets for different recognition tasks, the results show the effectiveness of the proposed search paradigm and demonstrate that the searched backbone is able to flexibly adapt different RSI recognition tasks and achieve impressive performance.

eess.IV↗

arXiv2021

Baby universes, ensemble averages and factorizations with matters

We investigate 2D topological gravity theories with matter fields turned on. We compute correlators of boundary creation operators with extra matter insertions. We provide a systematic procedure to determine a set of $α$-states on which the correlation functions of the different boundary operators factorize. Our results suggest that the boundary dual is an ensemble average of theories with different numbers of replicas of a seed matter contribution that is represented by excitations on top of the gravity sector. The construction of the $α$-states and the subsequent computations in the Hilbert space point of view are most conveniently done when the bulk couplings of the matter fields are expanded in a basis of vortices, which can be thought of as a 1-dimensional analogue of the SD-branes. We further demonstrate the existence of null states and the conditions imposed by reflection positivity. Moreover, in the CGS model with only disk and cylinder topologies, we map the baby universe creation and annihilation operators to the coupling constants in the one time SYK model, which allows us to match the wormhole and half-wormhole contributions in the CGS model explicitly to those identified in the SYK model. This explicitly realizes Coleman's initial idea. Some details of the computations in a variety of concrete models are presented.

hep-th↗

arXiv2021

GAN-based disentanglement learning for chest X-ray rib suppression

Clinical evidence has shown that rib-suppressed chest X-rays (CXRs) can improve the reliability of pulmonary disease diagnosis. However, previous approaches on generating rib-suppressed CXR face challenges in preserving details and eliminating rib residues. We hereby propose a GAN-based disentanglement learning framework called Rib Suppression GAN, or RSGAN, to perform rib suppression by utilizing the anatomical knowledge embedded in unpaired computed tomography (CT) images. In this approach, we employ a residual map to characterize the intensity difference between CXR and the corresponding rib-suppressed result. To predict the residual map in CXR domain, we disentangle the image into structure- and contrast-specific features and transfer the rib structural priors from digitally reconstructed radiographs (DRRs) computed by CT. Furthermore, we employ additional adaptive loss to suppress rib residue and preserve more details. We conduct extensive experiments based on 1,673 CT volumes, and four benchmarking CXR datasets, totaling over 120K images, to demonstrate that (i) our proposed RSGAN achieves superior image quality compared to the state-of-the-art rib suppression methods; (ii) combining CXR with our rib-suppressed result leads to better performance in lung disease classification and tuberculosis area detection.

eess.IV↗

arXiv2021

A 3d disordered superconformal fixed point

We initiate the study of a three dimensional disordered supersymmetric field theory. Specifically, we consider a $\mathcal{N}=2$ large $N$ Wess-Zumino like model with cubic superpotential involving couplings drawn from a Gaussian random ensemble. Taking inspiration from analyses of lower dimensional SYK like models we demonstrate that the theory flows to a strongly coupled superconformal fixed point in the infra-red. In particular, we obtain leading large $N$ spectral data and operator product coefficients at the critical point. Moreover, the analytic control accorded by the model allows us to compare our results against those derived in the conformal bootstrap program and demonstrate consistency with general expectations.

hep-th↗

arXiv2021

Production of neutron-rich heavy nuclei around N = 162 in multinucleon transfer reactions

Within the framework of the dinuclear system model, the production mechanism of neutron-rich heavy nuclei around N = 162 has been investigated systematically. The isotopic yields in the multinucleon transfer reaction of $^{238}$U + $^{248}$Cm was analyzed and compared the available experimental data. Systematics on the production of superheavy nuclei via $^{238}$U on $^{252,254}$Cf, $^{254}$Es and $^{257}$Fm is investigated. It is found that the shell effect is of importance in the formation of neutron-rich nuclei around N=162 owing to the enhancement of fission barrier. The fragments in the multinucleon transfer reactions manifest the broad isotopic distribution and are dependent on the beam energy. The polar angles of the fragments tend to the forward emission with increasing the beam energy. The production cross sections of new isotopes are estimated and heavier targets are available for the neutron-rich superheavy nucleus formation. The optimal system and beam energy are proposed for the future experimental measurements.

nucl-th↗

arXiv2021

Inverse Aerodynamic Design of Gas Turbine Blades using Probabilistic Machine Learning

One of the critical components in Industrial Gas Turbines (IGT) is the turbine blade. Design of turbine blades needs to consider multiple aspects like aerodynamic efficiency, durability, safety and manufacturing, which make the design process sequential and iterative.The sequential nature of these iterations forces a long design cycle time, ranging from several months to years. Due to the reactionary nature of these iterations, little effort has been made to accumulate data in a manner that allows for deep exploration and understanding of the total design space. This is exemplified in the process of designing the individual components of the IGT resulting in a potential unrealized efficiency. To overcome the aforementioned challenges, we demonstrate a probabilistic inverse design machine learning framework (PMI), to carry out an explicit inverse design. PMI calculates the design explicitly without excessive costly iteration and overcomes the challenges associated with ill-posed inverse problems. In this work, the framework will be demonstrated on inverse aerodynamic design of three-dimensional turbine blades.

eess.SP↗

arXiv2021

Borromean droplet in three-component ultracold Bose gases

Borromean ring refers to a peculiar structure where three rings are linked together while any two of them are unlinked. Here we propose the realization of its quantum mechanical analog in a many-body system of three-component ultracold bosons. Namely, we identify the {\it Borromean droplet}, where only the ternary bosons can form a self-bound droplet while any binary subsystems cannot. Its formation is facilitated by an additional attractive force induced by the density fluctuation of a third component, which enlarges the mean-field collapse region in comparison to the binary case and renders the formation of Borromean droplet after incorporating the repulsive force from quantum fluctuations. Outside the Borromean regime, the competition between ternary and binary droplets leads to an interesting phenomenon of droplet phase separation, manifested by double plateaus in the density profile. We further show that the transition between different droplets and gas phase can be conveniently tuned by boson numbers and interaction strengths. The study reveals the possibility of Borromean binding in the many-body world and sheds light on more intriguing many-body bound state formed in multi-component systems.

cond-mat.quant-gas↗

arXiv2021

Nature of polaron-molecule transition in Fermi polarons

In this work, we explore the polaron and molecule physics by utilizing a unified variational ansatz with up to two particle-hole(p-h) excitations(V-2ph). We confirm the existence of a first-order transition in 3D and 2D Fermi polarons, and show that the nature of such transition lies in an energy competition between systems with different momenta ${\mathbf Q}=0$ and $|{\mathbf Q}|=k_F$, here ${\mathbf Q}$ is defined as the momentum of Fermi polaron system with respect to the Fermi sea of majority fermions (with Fermi momentum $k_F$). The literally proposed molecule ansatz is identified as an asymptotic limit of $|{\mathbf Q}|=k_F$ state in strong coupling regime, which implies a huge $SO(3)$(for 3D) or $SO(2)$ (for 2D) ground state degeneracy in this regime. The recognization of such degeneracy is crucially important for evaluating the molecule occupation in realistic systems with finite impurity density and at finite temperature. To compare with recent experiment of 3D Fermi polarons, we have calculated various physical quantities under the V-2ph framework and obtained results that are in good agreements with experimental data in the weak coupling and near resonance regime. Further, to check the validity of our conclusion in 2D, we have adopted a different variational method based on the Gaussian sample of high-order p-h excitations(V-Gph), and found the same conclusion on the nature of polaron-molecule transition therein. For 1D system, the V-2ph method predicts no sharp transition and the ground state is always at ${\mathbf Q}=0$ sector, consistent with exact Bethe ansatz solution. The presence/absence of polaron-molecule transition is analyzed to be closely related to the interplay effect of Pauli-blocking and p-h excitations in different dimensions.

cond-mat.quant-gas↗

arXiv2021

Doping quantum spin liquids on the Kagome lattice

We review recent density-matrix renormalization group (DMRG) studies of lightly doped quantum spin liquids (QSLs) on the kagome lattice. While a number of distinct conducting phases, including high-temperature superconductivity, have been theoretically anticipated we find instead a tendency toward fractionalized insulating charge-density-wave (CDW) states. In agreement with earlier work (Jiang, Devereaux, and Kivelson, Phys. Rev. Lett. ${\bf 119}$, 067002 (2017)), results for the $t$-$J$ model reveal that starting from a fully gapped QSL, light doping leads to CDW long-range order with a pattern that depends on lattice geometry and doping concentration such that there is one doped-hole per CDW unit cell, while the spin-spin correlations remain short-ranged. Alternatively, this state can be viewed as a stripe crystal or Wigner crystal of spinless holons, rather than doped holes. From here, by studying generalized versions of the $t$-$J$ model, we extend these results to light doping of other types of QSLs, including critical and chiral QSLs. Our results suggest that doping these QSLs also leads to insulating states with long-range CDW orders. While the superconducting correlations are short-ranged, they can be significantly enhanced by second-neighbor electron hopping. The relevance of our numerical results to Kagome materials is also discussed.

cond-mat.str-el↗

arXiv2021

Gapless spin liquid and pair density wave of the Hubbard model on three-leg triangular cylinders

We study the ground state properties of the Hubbard model on three-leg triangular cylinders using large-scale density-matrix renormalization group simulations. At half-filling, we identify an intermediate gapless spin liquid phase between a metallic phase at weak coupling and Mott insulating dimer phase at strong interaction, which has one gapless spin mode and algebraic spin-spin correlations but exponential decay scalar chiral-chiral correlations. Upon light doping the gapless spin liquid, the system exhibits power-law charge-density-wave (CDW) correlations but short-range single-particle, spin-spin, and chiral-chiral correlations. Similar to CDW correlations, the superconducting correlations are also quasi-long-ranged but oscillate in sign as a function of distance, which is consistent with the striped pair-density wave. When further doping the gapless spin liquid phase or doping the dimer order phase, another phase takes over, which has similar CDW correlations but all other correlations decay exponentially.

cond-mat.str-el↗

arXiv2021

Entanglement and Confinement in Coupled Quantum Systems

We study some general properties of coupled quantum systems. We consider simple interactions between two copies of identical Hamiltonians such as the SYK model, Pauli spin chains with random magnetic field and harmonic oscillators. Such couplings make the ground states close to the thermofield double states of the uncoupled Hamiltonians. For the coupled SYK model, we push the numerical computation further towards the thermodynamic limit so that an extrapolation in the size of the system is possible. We find good agreement between the extrapolated numerical result and the analytic result in the large-$q$ limit. We also consider the coupled gauged matrix model and vector model, and argue that the deconfinement is associated with the loss of the entanglement, similarly to the previous observation for the coupled SYK model. The understanding of the microscopic mechanism of the confinement/deconfinement transition enables us to estimate the quantum entanglement precisely, and backs up the dual gravity interpretation which relates the deconfinement to the disappearance of the wormhole. Our results demonstrate the importance of the entanglement between the color degrees of freedom in the emergence of the bulk geometry from quantum field theory via holography.

hep-th↗

arXiv2021

U-DuDoNet: Unpaired dual-domain network for CT metal artifact reduction

Recently, both supervised and unsupervised deep learning methods have been widely applied on the CT metal artifact reduction (MAR) task. Supervised methods such as Dual Domain Network (Du-DoNet) work well on simulation data; however, their performance on clinical data is limited due to domain gap. Unsupervised methods are more generalized, but do not eliminate artifacts completely through the sole processing on the image domain. To combine the advantages of both MAR methods, we propose an unpaired dual-domain network (U-DuDoNet) trained using unpaired data. Unlike the artifact disentanglement network (ADN) that utilizes multiple encoders and decoders for disentangling content from artifact, our U-DuDoNet directly models the artifact generation process through additions in both sinogram and image domains, which is theoretically justified by an additive property associated with metal artifact. Our design includes a self-learned sinogram prior net, which provides guidance for restoring the information in the sinogram domain, and cyclic constraints for artifact reduction and addition on unpaired data. Extensive experiments on simulation data and clinical images demonstrate that our novel framework outperforms the state-of-the-art unpaired approaches.

eess.IV↗

arXiv2020

A thermodynamically consistent pseudo-potential lattice Boltzmann model for multi-component, multiphase, partially miscible mixtures

Current multi-component, multiphase pseudo-potential lattice Boltzmann models have thermodynamic inconsistencies that prevent them to correctly predict the thermodynamic phase behavior of partially miscible multi-component mixtures, such as hydrocarbon mixtures. This paper identifies these inconsistencies and attempts to design a thermodynamically consistent multi-component, multiphase pseudo-potential lattice Boltzmann model that allows mass transfer across the phase interfaces and is capable to predict the phase behavior of typically partially miscible hydrocarbon mixtures. The designed model defines the total interaction force for the entire phase and split the force into individual components. Through a properly derived force split factor associated with the volatility of each component, the model can achieve precise thermodynamic consistency in multi-component hydrocarbon mixtures, which is described by the iso-fugacity rule.

physics.comp-ph↗