Showing 1–15 of 15 results for author: Tang, W H

PICProp: Physics-Informed Confidence Propagation for Uncertainty Quantification

Authors: Qianli Shen, Wai Hoh Tang, Zhun Deng, Apostolos Psaros, Kenji Kawaguchi

Abstract: Standard approaches for uncertainty quantification in deep learning and physics-informed learning have persistent limitations. Indicatively, strong assumptions regarding the data likelihood are required, the performance highly depends on the selection of priors, and the posterior can be sampled only approximately, which leads to poor approximations because of the associated computational cost. Thi… ▽ More Standard approaches for uncertainty quantification in deep learning and physics-informed learning have persistent limitations. Indicatively, strong assumptions regarding the data likelihood are required, the performance highly depends on the selection of priors, and the posterior can be sampled only approximately, which leads to poor approximations because of the associated computational cost. This paper introduces and studies confidence interval (CI) estimation for deterministic partial differential equations as a novel problem. That is, to propagate confidence, in the form of CIs, from data locations to the entire domain with probabilistic guarantees. We propose a method, termed Physics-Informed Confidence Propagation (PICProp), based on bi-level optimization to compute a valid CI without making heavy assumptions. We provide a theorem regarding the validity of our method, and computational experiments, where the focus is on physics-informed learning. △ Less

Submitted 20 October, 2023; v1 submitted 10 October, 2023; originally announced October 2023.

Comments: Accepted at NeurIPS 2023. Code is available at https://github.com/ShenQianli/PICProp

MixupE: Understanding and Improving Mixup from Directional Derivative Perspective

Authors: Yingtian Zou, Vikas Verma, Sarthak Mittal, Wai Hoh Tang, Hieu Pham, Juho Kannala, Yoshua Bengio, Arno Solin, Kenji Kawaguchi

Abstract: Mixup is a popular data augmentation technique for training deep neural networks where additional samples are generated by linearly interpolating pairs of inputs and their labels. This technique is known to improve the generalization performance in many learning paradigms and applications. In this work, we first analyze Mixup and show that it implicitly regularizes infinitely many directional deri… ▽ More Mixup is a popular data augmentation technique for training deep neural networks where additional samples are generated by linearly interpolating pairs of inputs and their labels. This technique is known to improve the generalization performance in many learning paradigms and applications. In this work, we first analyze Mixup and show that it implicitly regularizes infinitely many directional derivatives of all orders. Based on this new insight, we propose an improved version of Mixup, theoretically justified to deliver better generalization performance than the vanilla Mixup. To demonstrate the effectiveness of the proposed method, we conduct experiments across various domains such as images, tabular data, speech, and graphs. Our results show that the proposed method improves Mixup across multiple datasets using a variety of architectures, for instance, exhibiting an improvement over Mixup by 0.8% in ImageNet top-1 accuracy. △ Less

Submitted 15 October, 2023; v1 submitted 27 December, 2022; originally announced December 2022.

Comments: 16 pages, Best Student Paper Award at UAI 2023

Interaction of a priori Anatomic Knowledge with Self-Supervised Contrastive Learning in Cardiac Magnetic Resonance Imaging

Authors: Makiya Nakashima, Inyeop Jang, Ramesh Basnet, Mitchel Benovoy, W. H. Wilson Tang, Christopher Nguyen, Deborah Kwon, Tae Hyun Hwang, David Chen

Abstract: Training deep learning models on cardiac magnetic resonance imaging (CMR) can be a challenge due to the small amount of expert generated labels and inherent complexity of data source. Self-supervised contrastive learning (SSCL) has recently been shown to boost performance in several medical imaging tasks. However, it is unclear how much the pre-trained representation reflects the primary organ of… ▽ More Training deep learning models on cardiac magnetic resonance imaging (CMR) can be a challenge due to the small amount of expert generated labels and inherent complexity of data source. Self-supervised contrastive learning (SSCL) has recently been shown to boost performance in several medical imaging tasks. However, it is unclear how much the pre-trained representation reflects the primary organ of interest compared to spurious surrounding tissue. In this work, we evaluate the optimal method of incorporating prior knowledge of anatomy into a SSCL training paradigm. Specifically, we evaluate using a segmentation network to explicitly local the heart in CMR images, followed by SSCL pretraining in multiple diagnostic tasks. We find that using a priori knowledge of anatomy can greatly improve the downstream diagnostic performance. Furthermore, SSCL pre-training with in-domain data generally improved downstream performance and more human-like saliency compared to end-to-end training and ImageNet pre-trained networks. However, introducing anatomic knowledge to pre-training generally does not have significant impact. △ Less

Submitted 24 May, 2022; originally announced May 2022.

Comments: Under review at Machine Learning in Healthcare

Quantum Spin Dynamics in a Normal Bose Gas with Spin-orbit Coupling

Authors: Wai Ho Tang, Shizhong Zhang

Abstract: In this Letter, we investigate spin dynamics of a two-component Bose gas with spin-orbit coupling realised in cold atom experiments. We derive coupled hydrodynamic equations for number and spin densities as well as their associated currents. Specialising to quasi-one-dimensional situation, we obtain analytic solutions of the spin helix structure and its dynamics in both adiabatic and diabatic regi… ▽ More In this Letter, we investigate spin dynamics of a two-component Bose gas with spin-orbit coupling realised in cold atom experiments. We derive coupled hydrodynamic equations for number and spin densities as well as their associated currents. Specialising to quasi-one-dimensional situation, we obtain analytic solutions of the spin helix structure and its dynamics in both adiabatic and diabatic regimes. In the adiabatic regime, the transverse spin decays parabolically in the short-time limit and exponentially in the long time limit, depending on initial polarisation. In contrast, in the diabatic regime, transverse spin density and current oscillate in a way similar to the charge-current oscillation in an undamped LC circuit. The effects of Rabi coupling on the short-time spin dynamics is also discussed. Finally, using realistic experimental parameters for $^{87}$Rb, we show that the time scales for spin dynamics is of order of milliseconds to a few seconds and can be observed experimentally. △ Less

Submitted 30 July, 2018; v1 submitted 3 June, 2018; originally announced June 2018.

Comments: 6+6 pages, 3 figures

Journal ref: Phys. Rev. Lett. 121, 120403 (2018)

Model identification for ARMA time series through convolutional neural networks

Authors: Wai Hoh Tang, Adrian Röllin

Abstract: In this paper, we use convolutional neural networks to address the problem of model identification for autoregressive moving average time series models. We compare the performance of several neural network architectures, trained on simulated time series, with likelihood based methods, in particular the Akaike and Bayesian information criteria. We find that our neural networks can significantly out… ▽ More In this paper, we use convolutional neural networks to address the problem of model identification for autoregressive moving average time series models. We compare the performance of several neural network architectures, trained on simulated time series, with likelihood based methods, in particular the Akaike and Bayesian information criteria. We find that our neural networks can significantly outperform these likelihood based methods in terms of accuracy and, by orders of magnitude, in terms of speed. △ Less

Submitted 18 July, 2020; v1 submitted 11 April, 2018; originally announced April 2018.

Comments: 17 pages, 1 figure, 4 tables

MSC Class: 62M10

Efficient excitation of a symmetric collective atomic state with a single-photon through dipole blockade

Authors: Fang-Yu Hong, Shi-Jie Xiong, W. H. Tang

Abstract: In the famous quantum communication scheme developed by Duan {\it et al.}[L.M. Duan, M.D. Lukin, J.I. Cirac, and P. Zoller, Nature (London) {\bf 414} 413 (2001)], the probability of successful generating a symmetric collective atomic state with a single-photon emitted have to be far smaller than 1 to obtain an acceptable entangled state. Because of strong dipole-dipole interaction between two Ry… ▽ More In the famous quantum communication scheme developed by Duan {\it et al.}[L.M. Duan, M.D. Lukin, J.I. Cirac, and P. Zoller, Nature (London) {\bf 414} 413 (2001)], the probability of successful generating a symmetric collective atomic state with a single-photon emitted have to be far smaller than 1 to obtain an acceptable entangled state. Because of strong dipole-dipole interaction between two Rydberg atoms, more than one simultaneous excitations in an atomic ensembles are greatly suppressed, which makes it possible to excite a mesoscopic cold atomic ensemble into a singly-excited symmetric collective state accompanied by a signal photon with near unity success probability, at the same higher-order excitations can be significantly inhibited. △ Less

Submitted 3 March, 2010; originally announced March 2010.

Comments: 4 pages, 1 figures

Long-distance quantum communication with "polarization" maximally entangled states

Authors: Fang-Yu Hong, Shi-Jie Xiong, W. H. Tang

Abstract: We propose a scheme for long-distance quantum communication where the elementary entanglement is generated through two-photon interference and quantum swap** is performed through one-photon interference. Local "polarization" maximally entangled states of atomic ensembles are generated by absorbing a single photon from on-demand single-photon sources. This scheme is robust against phase fluctua… ▽ More We propose a scheme for long-distance quantum communication where the elementary entanglement is generated through two-photon interference and quantum swap** is performed through one-photon interference. Local "polarization" maximally entangled states of atomic ensembles are generated by absorbing a single photon from on-demand single-photon sources. This scheme is robust against phase fluctuations in the quantum channels, moreover speeds up long-distance high-fidelity entanglement generation rate. △ Less

Submitted 3 March, 2010; v1 submitted 22 September, 2008; originally announced September 2008.

Comments: 5 pages 5 figures

Sphaleron rate at high temperature in 1+1 dimensions

Authors: W. H. Tang, J. Smit

Abstract: We resolve the controversy in the high temperature behavior of the sphaleron rate in the abelian Higgs model in 1+1 dimensions. The $T^2$ behavior at intermediate lattice spacings is found to change into $T^{2/3}$ behavior in the continuum limit. The results are supported by analytic arguments that the classical approximation is good for this model. We resolve the controversy in the high temperature behavior of the sphaleron rate in the abelian Higgs model in 1+1 dimensions. The $T^2$ behavior at intermediate lattice spacings is found to change into $T^{2/3}$ behavior in the continuum limit. The results are supported by analytic arguments that the classical approximation is good for this model. △ Less

Submitted 24 September, 1998; originally announced September 1998.

Comments: LATTICE98(electroweak)

Report number: ITFA-98-22

Journal ref: Nucl.Phys.Proc.Suppl. 73 (1999) 665-667

High Temperature Behavior of the Chern-Simons Diffusion Rate in the 1+1 D Abelian Higgs Model

Authors: W. H. Tang, J. Smit

Abstract: We give arguments that in the 1+1 dimensional abelian Higgs model the classical approximation can be good for the leading high temperature behavior of real time processes. The Chern-Simons diffusion rate (`sphaleron rate') is studied numerically in this approximation. New results at high temperature show a $T^{2/3}$ behavior of the rate at sufficiently small lattice spacing. We give arguments that in the 1+1 dimensional abelian Higgs model the classical approximation can be good for the leading high temperature behavior of real time processes. The Chern-Simons diffusion rate (`sphaleron rate') is studied numerically in this approximation. New results at high temperature show a $T^{2/3}$ behavior of the rate at sufficiently small lattice spacing. △ Less

Submitted 22 September, 1998; v1 submitted 4 May, 1998; originally announced May 1998.

Comments: Corrected trivial errors, extended conclusions somewhat

Report number: ITFA-98-08

Journal ref: Nucl.Phys. B540 (1999) 437-456

Theory for the nonequilibrium dynamics of flexible chain molecules: relaxation to equilibrium of pentadecane from an all-trans conformation

Authors: Wilfred H. Tang, Konstantin S. Kostov, Karl F. Freed

Abstract: We extend to nonequilibrium processes our recent theory for the long time dynamics of flexible chain molecules. While the previous theory describes the equilibrium motions for any bond or interatomic separation in (bio)polymers by time correlation functions, the present extension of the theory enables the prediction of the nonequilibrium relaxation that occurs in processes, such as T-jump experi… ▽ More We extend to nonequilibrium processes our recent theory for the long time dynamics of flexible chain molecules. While the previous theory describes the equilibrium motions for any bond or interatomic separation in (bio)polymers by time correlation functions, the present extension of the theory enables the prediction of the nonequilibrium relaxation that occurs in processes, such as T-jump experiments, where there are sudden transitions between, for example, different equilibrium states. As a test of the theory, we consider the ``unfolding'' of pentadecane when it is transported from a constrained all-trans conformation to a random-coil state at thermal equilibrium. The time evolution of the mean-square end-to-end distance after release of the constraint is computed both from the theory and from Brownian dynamics (BD) simulations. The predictions of the theory agree very well with the BD simulations. Furthermore, the theory produces enormous savings in computer time. This work is a starting point for the application of the new method to nonequilibrium processes with biological importance such as the helix-coil transition and protein folding. △ Less

Submitted 25 February, 1998; originally announced February 1998.

Comments: 11 pages total, including 2 Postscript figures; submitted to Journal of Chemical Physics

Quenched degrees of freedom in symmetric diblock copolymer thin films

Authors: Wilfred H. Tang, Thomas A. Witten

Abstract: We study the effect of monomer immobilization (quenching) on the orientation of the lamellae in symmetric diblock copolymer thin films with neutrally wetting surfaces. A small fraction of the monomers immediately next to the solid substrate is presumed to be quenched. In both the weak segregation limit and the strong segregation limit, quenching favors the lamellae orienting perpendicular to the… ▽ More We study the effect of monomer immobilization (quenching) on the orientation of the lamellae in symmetric diblock copolymer thin films with neutrally wetting surfaces. A small fraction of the monomers immediately next to the solid substrate is presumed to be quenched. In both the weak segregation limit and the strong segregation limit, quenching favors the lamellae orienting perpendicular to the film. Quenching inhibits the order-disorder transition twice as much for the parallel orientation as for the perpendicular. △ Less

Submitted 5 December, 1997; originally announced December 1997.

Comments: 11 pages

Numerical study of plasmon properties in the SU(2)-Higgs model

Authors: W. H. Tang, J. Smit

Abstract: Using the (effective) classical approximation, we compute numerically time-dependent correlation functions in the SU(2)-Higgs model around the electroweak phase transition, for $m_H \approx m_W$. The parameters of the classical model have been determined previously by the dimensional reduction relations for time-independent correlators. The $H$ and $W$ correlation functions correspond to gauge i… ▽ More Using the (effective) classical approximation, we compute numerically time-dependent correlation functions in the SU(2)-Higgs model around the electroweak phase transition, for $m_H \approx m_W$. The parameters of the classical model have been determined previously by the dimensional reduction relations for time-independent correlators. The $H$ and $W$ correlation functions correspond to gauge invariant fields. They show damped oscillatory behavior from which we extract frequencies $\om$ and dam** rates $\gm$. In the Higgs phase the dam** rates have roughly the values obtained in analytic calculations in the quantum theory. In the plasma phase (where analytic estimates for gauge invariant fields are not available), the dam** rate associated with $H$ is an order of magnitude larger than in the Higgs phase, while the $W$ correlator appears to be overdamped, with a small rate. The frequency $\om_H$ shows a clear dip at the transition. The results are approximately independent of the lattice spacing, but this appears to be compatible with the lattice spacing dependence expected from perturbation theory. △ Less

Submitted 18 August, 1997; v1 submitted 20 February, 1997; originally announced February 1997.

Comments: 27 pages, latex, 13 figures. Improved presentation, data unchanged. New is the conclusion that the apparent lattice spacing independence of the `plasmon' frequency is compatible with the divergencies expected from perturbation theory. (Version resubmitted to NPB on 17 July 1997.)

Report number: ITFA-97-02

Journal ref: Nucl.Phys. B510 (1998) 401-420

Chern-Simons Diffusion Rate near the Electroweak Phase Transition for $m_H \approx m_W$

Authors: Jan Smit, Wai Hung Tang

Abstract: The rate of $B$-violation in the standard model at finite temperature is closely related to the diffusion rate $Γ$ of Chern-Simons number. We compute this rate for $m_H \approx m_W$ in the classical approximation in an effective SU(2)-Higgs model, using Krasnitz's algorithm. The parameters in the effective hamiltonian are determined by comparison with dimensional reduction. In the high temperatu… ▽ More The rate of $B$-violation in the standard model at finite temperature is closely related to the diffusion rate $Γ$ of Chern-Simons number. We compute this rate for $m_H \approx m_W$ in the classical approximation in an effective SU(2)-Higgs model, using Krasnitz's algorithm. The parameters in the effective hamiltonian are determined by comparison with dimensional reduction. In the high temperature phase we find $Γ/V (α_W T)^4 \approx 1$, neglecting a finite renormalization. In the low temperature phase near the transition we find the rate to be much larger than might be expected from previous analytic calculations based on the sphaleron. △ Less

Submitted 13 May, 1996; originally announced May 1996.

Comments: 23 pages, 8 Postscript figures, uses epsf.sty

Report number: ITFA-95-11

Journal ref: Nucl.Phys. B482 (1996) 265-285

Sphaleron transition rate at high temperature in the 1+1 D abelian Higgs model

Authors: Jan Smit, Wai Hung Tang

Abstract: New results for the rate are presented using the canonical ensemble in the classical approximation on a spatial lattice. We find that the rate at high temperatures is proportional to $T^2$, and strongly dependent on the lattice spacing $a$. We conclude that a better effective action is needed for the classical approximation. New results for the rate are presented using the canonical ensemble in the classical approximation on a spatial lattice. We find that the rate at high temperatures is proportional to $T^2$, and strongly dependent on the lattice spacing $a$. We conclude that a better effective action is needed for the classical approximation. △ Less

Submitted 2 December, 1994; originally announced December 1994.

Comments: 3 pages, uuencoded compressed postscript file, contribution to the Conference Proceedings "LATTICE 94", Bielefeld- Germany

Report number: ITFA-94-37

Journal ref: Nucl.Phys.Proc.Suppl. 42 (1995) 590-592

Sphaleron transition rate in the classical 1+1 dimensional abelian Higgs model at finite temperature

Authors: J. Smit, W. H. Tang

Abstract: We compute the sphaleron transition rate in the 1+1 dimensional abelian Higgs model at finite temperature, by real time simulation using the classical canonical ensemble. We compute the sphaleron transition rate in the 1+1 dimensional abelian Higgs model at finite temperature, by real time simulation using the classical canonical ensemble. △ Less

Submitted 2 December, 1993; originally announced December 1993.

Comments: 3 pages to appear in the Proceedings of Lattice '93, Dallas, Texas, 12-16 October 1993, comes as a single postscript file (LaTeX source available from the authors), ITFA 93-34

Journal ref: Nucl.Phys.Proc.Suppl. 34 (1994) 616-618