Academia.eduAcademia.edu

Outline

Title
Abstract
Training and Inference Procedure
Computational Cost Analysis
Analysis for PBMG-Xy
Analysis for PBMG-Φ 4
Conclusion
References
All Topics
Physics
Mathematical Physics

Scalable Lattice Sampling using Factorized Generative Models

Profile image of ankur singhaankur singha

2023, arXiv (Cornell University)

https://doi.org/10.48550/ARXIV.2308.08615
visibility

…

description

29 pages

link

1 file

Abstract

Boltzmann distributions over lattices are pervasive in Computational Physics. Sampling them becomes increasingly difficult with increasing lattice-size, especially near critical regions, e.g., phase transitions in statistical systems and continuum limits in lattice field theory. Machine learning-based methods, such as normalizing flows, have demonstrated the ability to alleviate these issues for small lattices. However, scaling to large lattices is a major challenge. We present a novel approach called Parallelizable Block Metropolis-within-Gibbs (PBMG) for generating samples for any lattice model. It factorizes the joint distribution of the lattice into local parametric kernels, thereby allowing efficient sampling of very large lattices. We validate our approach on the XY model and the Scalar ϕ 4 theory. PBMG achieves high acceptance rates and less correlated samples, and the observable statistics estimated from the samples match the ground truth. Moreover, PBMG significantly speeds up inference for large lattices as compared to HMC and plain MCMC algorithms.

Related papers

Coupled Coarse Graining and Markov Chain Monte Carlo for Lattice Systems
E. Kalligiannaki

Lecture Notes in Computational Science and Engineering, 2011

We propose an efficient Markov Chain Monte Carlo method for sampling equilibrium distributions for stochastic lattice models, capable of handling correctly long and short-range particle interactions. The proposed method is a Metropolistype algorithm with the proposal probability transition matrix based on the coarsegrained approximating measures introduced in . We prove that the proposed algorithm reduces the computational cost due to energy differences and has comparable mixing properties with the classical microscopic Metropolis algorithm, controlled by the level of coarsening and reconstruction procedure. The properties and effectiveness of the algorithm are demonstrated with an exactly solvable example of a one dimensional Ising-type model, comparing efficiency of the single spin-flip Metropolis dynamics and the proposed coupled Metropolis algorithm.

View PDFchevron_right
On Estimation of Thermodynamic Observables in Lattice Field Theories with Deep Generative Models
paolo stornati

2020

In this work, we demonstrate that applying deep generative machine learning models for lattice field theory is a promising route for solving problems where Markov Chain Monte Carlo (MCMC) methods are problematic. More specifically, we show that generative models can be used to estimate the absolute value of the free energy, which is in contrast to existing MCMC-based methods which are limited to only estimate free energy differences. We demonstrate the effectiveness of the proposed method for two-dimensional ϕ^4 theory and compare it to MCMC-based methods in detailed numerical experiments.

View PDFchevron_right
lbmpy: Automatic code generation for efficient parallel lattice Boltzmann methods
Ulrich Rüde

Journal of Computational Science, 2021

Lattice Boltzmann methods are a popular mesoscopic alternative to classical computational fluid dynamics based on the macroscopic equations of continuum mechanics. Many variants of lattice Boltzmann methods have been developed that vary in complexity, accuracy, and computational cost. Extensions are available to simulate multi-phase, multi-component, turbulent, and non-Newtonian flows. In this work we present lbmpy, a code generation package that supports a wide variety of different lattice Boltzmann methods. Additionally, lbmpy provides a generic development environment for new schemes. A high-level domainspecific language allows the user to formulate, extend and test various lattice Boltzmann methods. In all cases, the lattice Boltzmann method can be specified in symbolic form. Transformations that operate on this symbolic representation yield highly efficient compute kernels. This is achieved by automatically parallelizing the methods, and by various application-specific automatized steps that optimize the resulting code. This pipeline of transformations can be applied to a wide range of lattice Boltzmann variants, including single-and two-relaxation-time schemes, multi-relaxationtime methods, as well as the more advanced cumulant methods, and entropically stabilized methods. lbmpy can be integrated into high-performance computing frameworks to enable massively parallel, distributed simulations. This is demonstrated using the WALBERLA multiphysics package to conduct scaling experiments on the SuperMUC-NG supercomputing system on up to 147 456 compute cores.

View PDFchevron_right
Factorization symmetry in lattice Boltzmann simulations
Pietro Asinari

2009

A non-perturbative algebraic theory of lattice Boltzmann method is developed based on a symmetry of a product. It involves three steps: (i) Derivation of admissible lattices in one spatial dimension through a matching condition which imposes restricted extension of higher-order Gaussian moments, (ii) Special quasi-equilibrium distribution function found analytically in closed form on the product-lattice in two and three spatial dimensions, and which proves factorization of quasi-equilibrium moments, and (iii) Algebraic method of pruning based on a one-into-one relation between groups of discrete velocities and moments. Two routes of constructing lattice Boltzmann equilibria are distinguished. Present theory includes previously known limiting and special cases of lattices, and enables automated derivation of lattice Boltzmann models from two-dimensional tables, by finding roots of one polynomial and solving a few linear systems.

View PDFchevron_right
Towards Exascale Lattice Boltzmann computing
Giorgio Amati

Computers & Fluids, 2019

1 D3Q27 lattice, composed of a set of 27 discrete velocities (left). D3Q19 lattice, composed of a set of 19 discrete velocities (right). 2 Roofline Model obtained for an intel Phi processor: for F/B < 10 the effective processing speed grows linearly with F/B, whereas for F/B > 10 it comes to a saturation, dictated by the raw processing rate. Increasing F/B much beyond, does not win any additional performance. The vertical line indicates the performance range for a "Vanilla Lattice Boltzmann".

View PDFchevron_right
Lattice Boltzmann computations and applications to physics
Bastien Chopard

Theoretical Computer Science, 1999

We discuss the lattice Boltzmann computing approach, its connection with cellular automata and present a new mode1 for simulating wave propagation in complex environments. We illustrate the behavior of our mode1 on several applications like radio wave propagation in a city, solid body motion and fracture phenomena.

View PDFchevron_right
Challenges in lattice Boltzmann computing
Sauro Succi

Journal of Statistical Physics, 1995

Some of the most urgent challenges facing the lattice Boltzmann equation (LBE) to rival state-of-the-art computer fluid dynamics (CFD) techniques are discussed. A novel LBE scheme fork-ɛ turbulence modeling is proposed.

View PDFchevron_right
Thread-Safe Lattice Boltzmann for High-Performance Computing on GPUs
Giorgio Amati

2023

We present thread-safe, highly-optimized lattice Boltzmann implementations, specifically aimed at exploiting the high memory bandwidth of GPUbased architectures. At variance with standard approaches to LB coding, the proposed strategy, based on the reconstruction of the post-collision distribution via Hermite projection, enforces data locality and avoids the onset of memory dependencies, which may arise during the propagation step, with no need to resort to more complex streaming strategies. The thread-safe lattice Boltzmann achieves peak performances, both in two and three dimensions and it allows to sensibly reduce the allocated memory (tens of GigaBytes for order billions lattice nodes simulations) by retaining the algorithmic simplicity of standard LB computing. Our findings open attractive prospects for high-performance simulations of complex flows on GPU-based architectures.

View PDFchevron_right
Entropic lattice Boltzmann methods
Bruce Boghosian

Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 2001

We present a general methodology for constructing lattice Boltzmann models of hydrodynamics with certain desired features of statistical physics and kinetic theory. We show how a methodology of linear programming theory, known as Fourier-Motzkin elimination, provides an important tool for visualizing the state space of lattice Boltzmann algorithms that conserve a given set of moments of the distribution function. We show how such models can be endowed with a Lyapunov functional, analogous to Boltzmann's H, resulting in unconditional numerical stability. Using the Chapman-Enskog analysis and numerical simulation, we demonstrate that such entropically stabilized lattice Boltzmann algorithms, while fully explicit and perfectly conservative, may achieve remarkably low values for transport coefficients, such as viscosity. Indeed, the lowest such attainable values are limited only by considerations of accuracy, rather than stability. The method thus holds promise for high-Reynolds number simulations of the Navier-Stokes equations.

View PDFchevron_right
Fast automatic Bayesian cubature using lattice sampling
Fred J Hickernell

Statistics and Computing, 2019

Automatic cubatures approximate integrals to user-specified error tolerances. For high dimensional problems, it is difficult to adaptively change the sampling pattern, but one can automatically determine the sample size, n, given a reasonable, fixed sampling pattern. We take this approach here using a Bayesian perspective. We postulate that the integrand is an instance of a Gaussian stochastic process parameterized by a constant mean and a covariance kernel defined by a scale parameter times a parameterized function specifying how the integrand values at two different points in the domain are related. These hyperparameters are inferred or integrated out using integrand values via one of three techniques: empirical Bayes, full Bayes, or generalized cross-validation. The sample size, n, is increased until the half-width of the credible interval for the Bayesian posterior mean is no greater than the error tolerance. The process outlined above typically requires a computational cost of O(N opt n 3), where N opt is the number of optimization steps required to identify the hyperparameters. Our innovation is to pair low discrepancy nodes with matching covariance kernels to lower the computational cost to O(N opt n log n). This approach is demonstrated explicitly with rank-1 lattice sequences and shift-invariant kernels. Our algorithm is implemented in the Guaranteed Automatic Integration Library (GAIL).

View PDFchevron_right

References (50)

  1. Metropolis, N., Rosenbluth, A.W., Rosenbluth, M.N., Teller, A.H., Teller, E.: Equation of state calculations by fast computing machines. The journal of chemical physics 21(6), 1087-1092 (1953)
  2. Hastings, W.K.: Monte carlo sampling methods using markov chains and their applications (1970)
  3. Wolff, U.: Critical slowing down. Nuclear Physics B -Proceedings Supplements 17, 93-102 (1990) https://doi.org/10.1016/0920-5632(90)90224-I
  4. Schaefer, S., Sommer, R., Virotta, F., Collaboration, A., et al.: Critical slowing down and error analysis in lattice qcd simulations. Nuclear Physics B 845(1), 93-119 (2011)
  5. Swendsen, R.H., Wang, J.-S.: Nonuniversal critical dynamics in monte carlo simu- lations. Phys. Rev. Lett. 58, 86-88 (1987) https://doi.org/10.1103/PhysRevLett. 58.86
  6. Wolff, U.: Collective monte carlo updating for spin systems. Phys. Rev. Lett. 62, 361-364 (1989) https://doi.org/10.1103/PhysRevLett.62.361
  7. Prokof'ev, N.V., Svistunov, B.V., Tupitsyn, I.S.: "worm" algorithm in quantum monte carlo simulations. Physics Letters A 238(4), 253-257 (1998) https://doi. org/10.1016/S0375-9601(97)00957-2
  8. Evertz, H.G., Lana, G., Marcu, M.: Cluster algorithm for vertex models. Phys. Rev. Lett. 70, 875-879 (1993) https://doi.org/10.1103/PhysRevLett.70.875
  9. Syljuåsen, O.F., Sandvik, A.W.: Quantum monte carlo with directed loops. Phys. Rev. E 66, 046701 (2002) https://doi.org/10.1103/PhysRevE.66.046701
  10. Neal, R.M., et al.: Mcmc using hamiltonian dynamics. Handbook of markov chain monte carlo 2(11), 2 (2011)
  11. Cranmer, K., Kanwar, G., Racanière, S., Rezende, D.J., Shanahan, P.E.: Advances in machine-learning-based sampling motivated by lattice quantum chro- modynamics. Nature Rev. Phys. 5(9), 526-535 (2023) https://doi.org/10.1038/ s42254-023-00616-w arXiv:2309.01156 [hep-lat]
  12. Medvidovic, M., Carrasquilla, J., Hayward, L.E., Kulchytskyy, B.: Generative models for sampling of lattice field theories (2021)
  13. Albergo, M.S., Kanwar, G., Racanière, S., Rezende, D.J., Urban, J.M., Boyda, D., Cranmer, K., Hackett, D.C., Shanahan, P.E.: Flow-based sampling for fermionic lattice field theories. Phys. Rev. D 104(11), 114507 (2021) https://doi.org/10. 1103/PhysRevD.104.114507 arXiv:2106.05934 [hep-lat]
  14. Albergo, M.S., Boyda, D., Cranmer, K., Hackett, D.C., Kanwar, G., Racanière, S., Rezende, D.J., Romero-López, F., Shanahan, P.E., Urban, J.M.: Flow-based sampling in the lattice Schwinger model at criticality. Phys. Rev. D 106(1), 014514 (2022) https://doi.org/10.1103/PhysRevD.106.014514 arXiv:2202.11712 [hep-lat]
  15. Hackett, D.C., Hsieh, C.-C., Albergo, M.S., Boyda, D., Chen, J.-W., Chen, K.-F., Cranmer, K., Kanwar, G., Shanahan, P.E.: Flow-based sampling for multimodal distributions in lattice field theory. arXiv:2107.00734 (2021) arXiv:2107.00734 [hep-lat]
  16. Nicoli, K.A., Nakajima, S., Strodthoff, N., Samek, W., Müller, K.-R., Kessel, P.: Asymptotically unbiased estimation of physical observables with neural samplers. Phys. Rev. E 101, 023304 (2020) https://doi.org/10.1103/PhysRevE.101.023304
  17. Nicoli, K.A., Anders, C.J., Funcke, L., Hartung, T., Jansen, K., Kessel, P., Nakajima, S., Stornati, P.: Estimation of thermodynamic observables in lattice field theories with deep generative models. Phys. Rev. Lett. 126, 032001 (2021) https://doi.org/10.1103/PhysRevLett.126.032001
  18. Caselle, M., Cellini, E., Nada, A.: Sampling the lattice Nambu-Goto string using Continuous Normalizing Flows (2023) arXiv:2307.01107 [hep-lat]
  19. Reyes-Gonzalez, H., Torre, R.: The NFLikelihood: an unsupervised DNNLikeli- hood from Normalizing Flows (2023) arXiv:2309.09743 [hep-ph]
  20. Coccaro, A., Letizia, M., Reyes-Gonzalez, H., Torre, R.: On the curse of dimensionality for Normalizing Flows (2023) arXiv:2302.12024 [stat.ML]
  21. Kanwar, G., Albergo, M.S., Boyda, D., Cranmer, K., Hackett, D.C., Racanière, S., Rezende, D.J., Shanahan, P.E.: Equivariant flow-based sampling for lattice gauge theory. Physical Review Letters 125(12) (2020) https://doi.org/10.1103/ physrevlett.125.121601
  22. Eichhorn, T., Fuwa, G., Hoelbling, C., Varnhorst, L.: Parallel Tempered Meta- dynamics: Overcoming potential barriers without surfing or tunneling (2023) arXiv:2307.04742 [hep-lat]
  23. Abbott, R., et al.: Gauge-equivariant flow models for sampling in lattice field theories with pseudofermions. Phys. Rev. D 106(7), 074506 (2022) https://doi. org/10.1103/PhysRevD.106.074506 arXiv:2207.08945 [hep-lat]
  24. Abbott, R., et al.: Sampling QCD field configurations with gauge-equivariant flow models. PoS LATTICE2022, 036 (2023) https://doi.org/10.22323/1.430.0036 arXiv:2208.03832 [hep-lat]
  25. Abbott, R., et al.: Normalizing flows for lattice gauge theory in arbitrary space- time dimension (2023) arXiv:2305.02402 [hep-lat]
  26. Wang, L., Aarts, G., Zhou, K.: Diffusion Models as Stochastic Quantization in Lattice Field Theory (2023) arXiv:2309.17082 [hep-lat]
  27. Foreman, S., Izubuchi, T., Jin, L., Jin, X.-Y., Osborn, J.C., Tomiya, A.: HMC with Normalizing Flows. PoS LATTICE2021, 073 (2022) https://doi.org/10. 22323/1.396.0073 arXiv:2112.01586 [cs.LG]
  28. Foreman, S., Jin, X.-Y., Osborn, J.C.: LeapfrogLayers: A Trainable Framework for Effective Topological Sampling. PoS LATTICE2021, 508 (2022) https://doi. org/10.22323/1.396.0508 arXiv:2112.01582 [hep-lat]
  29. Boyda, D., et al.: Applications of Machine Learning to Lattice Quantum Field Theory. In: Snowmass 2021 (2022)
  30. Song, J., Zhao, S., Ermon, S.: A-NICE-MC: Adversarial Training for MCMC. arXiv (2017). https://doi.org/10.48550/ARXIV.1706.07561 . https://arxiv.org/ abs/1706.07561
  31. Liu, J., Qi, Y., Meng, Z.Y., Fu, L.: Self-learning monte carlo method. Phys. Rev. B 95, 041101 (2017) https://doi.org/10.1103/PhysRevB.95.041101
  32. Chen, C., Xu, X.Y., Liu, J., Batrouni, G., Scalettar, R., Meng, Z.Y.: Symmetry- enforced self-learning monte carlo method applied to the holstein model. Phys. Rev. B 98, 041102 (2018) https://doi.org/10.1103/PhysRevB.98.041102
  33. Huang, L., Wang, L.: Accelerated monte carlo simulations with restricted boltzmann machines. Physical Review B 95(3), 035105 (2017)
  34. Singh, J., Scheurer, M.S., Arora, V.: Conditional generative models for sampling and phase transition indication in spin systems. SciPost Phys. 11, 043 (2021) https://doi.org/10.21468/SciPostPhys.11.2.043
  35. Singha, A., Chakrabarti, D., Arora, V.: Generative learning for the problem of critical slowing down in lattice Gross-Neveu model. SciPost Phys. Core 5, 052 (2022) https://doi.org/10.21468/SciPostPhysCore.5.4.052 arXiv:2111.00574 [hep-lat]
  36. Pawlowski, J.M., Urban, J.M.: Reducing autocorrelation times in lattice sim- ulations with generative adversarial networks. Machine Learning: Science and Technology 1(4), 045011 (2020)
  37. Cristoforetti, M., Jurman, G., Nardelli, A.I., Furlanello, C.: Towards meaningful physics from generative models. arXiv (2017). https://doi.org/10.48550/ARXIV. 1705.09524 . https://arxiv.org/abs/1705.09524
  38. Noé, F., Olsson, S., Köhler, J., Wu, H.: Boltzmann generators: Sam- pling equilibrium states of many-body systems with deep learning. Science 365(6457), 1147 (2019) https://doi.org/10.1126/science.aaw1147 https://www.science.org/doi/pdf/10.1126/science.aaw1147
  39. Albergo, M.S., Kanwar, G., Shanahan, P.E.: Flow-based generative models for markov chain monte carlo in lattice field theory. Physical Review D 100(3), 034515 (2019)
  40. Singha, A., Chakrabarti, D., Arora, V.: Conditional normalizing flow for Markov chain Monte Carlo sampling in the critical region of lattice field theory. Phys. Rev. D 107(1), 014512 (2023) https://doi.org/10.1103/PhysRevD.107.014512
  41. Singha, A., Chakrabarti, D., Arora, V.: Sampling U (1) gauge theory using a re-trainable conditional flow-based model (2023) arXiv:2306.00581 [hep-lat]
  42. Abbott, R., Albergo, M.S., Botev, A., Boyda, D., Cranmer, K., Hackett, D.C., Matthews, A.G.D.G., Racanière, S., Razavi, A., Rezende, D.J., Romero-López, F., Shanahan, P.E., Urban, J.M.: Aspects of scaling and scalability for flow-based sampling of lattice QCD (2022)
  43. Wang, T., Wu, Y., Moore, D., Russell, S.J.: Meta-learning mcmc proposals. Advances in neural information processing systems 31 (2018)
  44. Durkan, C., Bekasov, A., Murray, I., Papamakarios, G.: Neural spline flows. Advances in neural information processing systems 32 (2019)
  45. Jones, G.L., Roberts, G.O., Rosenthal, J.S.: Convergence of conditional Metropolis-Hastings samplers. Advances in Applied Probability 46(2), 422-445 (2014)
  46. De, A.K., Harindranath, A., Maiti, J., Sinha, T.: Investigations in 1 + 1 dimen- sional lattice ϕ 4 theory. Phys. Rev. D 72, 094503 (2005) https://doi.org/10.1103/ PhysRevD.72.094503
  47. De, A.K., Harindranath, A., Maiti, J., Sinha, T.: Topological charge in 1 + 1 dimensional lattice ϕ 4 theory. Phys. Rev. D 72, 094504 (2005) https://doi.org/ 10.1103/PhysRevD.72.094504
  48. Vierhaus, I.: Simulation of phi 4 theory in the strong coupling expansion beyond the ising limit. Master's thesis, Humboldt-Universität zu Berlin, Mathematisch- Naturwissenschaftliche Fakultät I (2010). https://doi.org/10.18452/14138
  49. Vats, D., Flegal, J.M., Jones, G.L.: Multivariate output analysis for Markov chain Monte Carlo. Biometrika 106(2), 321-337 (2019) https: //doi.org/10.1093/biomet/asz002 https://academic.oup.com/biomet/article- pdf/106/2/321/28575440/asz002.pdf
  50. Albergo, M.S., Boyda, D., Hackett, D.C., Kanwar, G., Cranmer, K., Racanière, S., Rezende, D.J., Shanahan, P.E.: Introduction to Normalizing Flows for Lattice Field Theory (2021)

Related papers

Conditional normalizing flow for Markov chain Monte Carlo sampling in the critical region of lattice field theory
ankur singha

Physical Review D

The traditional Markov chain Monte Carlo (MCMC) suffers from the problem of critical slowing down. Generative machine learning methods, such as normalizing flows, offer a promising method to speed up MCMC simulations, especially in critical regions of lattice field theory. However, training these models for different parameter values in the critical region is inefficient. In this paper, we address this issue by interpolating or extrapolating the flow model to any parameter value in the critical region. We demonstrate the effectiveness of the proposed method for MCMC sampling in critical regions.

View PDFchevron_right
Conditional Normalizing flow for Monte Carlo sampling in lattice scalar field theory
ankur singha

2022

The cost of Monte Carlo sampling of lattice configurations is very high in the critical region of lattice field theory due to the high correlation between the samples. This paper suggests a Conditional Normalizing Flow (C-NF) model for sampling lattice configurations in the critical region to solve the problem of critical slowing down. We train the C-NF model using samples generated by Hybrid Monte Carlo (HMC) in non-critical regions with low simulation costs. The trained C-NF model is employed in the critical region to build a Markov chain of lattice samples with negligible autocorrelation. The C-NF model is used for both interpolation and extrapolation to the critical region of lattice theory. Our proposed method is assessed using the 1+1-dimensional scalar φ 4 theory. This approach enables the construction of lattice ensembles for many parameter values in the critical region, which reduces simulation costs by avoiding the critical slowing down.

View PDFchevron_right
The Lattice-Boltzmann method on optimal sampling lattices
Torsten Möller

2009

Abstract In this paper, we extend the single relaxation time lattice-Boltzmann method (LBM) to the 3D body-centered cubic (BCC) lattice. We show that the D3bQ15 lattice defined by a 15 neighborhood connectivity of the BCC lattice is not only capable of more accurately discretizing the velocity space of the continuous Boltzmann equation as compared to the D3Q15 Cartesian lattice, it also achieves a comparable spatial discretization with 30 percent less samples.

View PDFchevron_right
WaLBerla: Exploiting Massively Parallel Systems for Lattice Boltzmann Simulations
U. Rüde

Parallel Computing, 2009

transmitted, in any form or by any means, with the prior permission in writing of the publishers, or in the case of reprographic reproduction in accordance with the terms of licenses issued by the Copyright Licensing Agency. Enquiries concerning reproduction outside those terms should be sent to the publishers. The use of registered names, trademarks, etc., in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant laws and regulations and therefore free for general use. The publisher makes no representation, express or implied, with regard to the accuracy of the information contained in this book and cannot accept any legal responsibility or liability for any errors or omissions that may be made.

View PDFchevron_right
Lightweight Lattice Boltzmann
Giorgio Amati

arXiv (Cornell University), 2023

A GPU-accelerated version of the lattice Boltzmann method for efficient simulation of soft materials is introduced. Unlike standard approaches, this method reconstructs the distribution functions from available hydrodynamic variables (density, momentum, and pressure tensor) without storing the full set of discrete populations. This scheme shows satisfactory numerical stability, significantly lower memory requirements, and data access cost. A series of benchmark tests of relevance to soft matter, such as collisions of fluid droplets, is discussed to validate the method. The results can be of particular interest for high-performance simulations of soft matter systems on future exascale computers.

View PDFchevron_right

Related topics

  • Mathematics
  • Mathematical Physics
  • Computer Science
  • Artificial Intelligence
  • Physics
  • Acoustics
  • Quantum Mechanics
  • Lattice Field Theory
  • Geometry
  • Gauge theory
  • Gibbs sampling
    • 580 California St., Suite 400
    San Francisco, CA, 94104
    © 2025 Academia. All rights reserved