{"type":"ScholarlyArticle","title":"DPVI: A Dynamic-Weight Particle-Based Variational Inference Framework","authors":[{"name":"Chao Zhang","slug":"chao-zhang","given":"Chao","family":"Zhang","affiliation":"College of Computer Science and Technology , Zhejiang University; Alibaba-Zhejiang University Joint Research Institute of Frontier Technologies"},{"name":"Zhijian Li","slug":"zhijian-li","given":"Zhijian","family":"Li","affiliation":"Information Science and Electronic Engineering , Zhejiang University"},{"name":"Xin Du","slug":"xin-du","given":"Xin","family":"Du","affiliation":"Information Science and Electronic Engineering , Zhejiang University"},{"name":"Hui Qian","slug":"hui-qian","given":"Hui","family":"Qian","affiliation":"College of Computer Science and Technology , Zhejiang University; Alibaba-Zhejiang University Joint Research Institute of Frontier Technologies"}],"abstract":"The recently developed Particle-based Variational Inference (ParVI) methods drive the empirical distribution of a set of fixed-weight particles towards a given target distribution by iteratively updating particles' positions. However, the fixed weight restriction greatly confines the empirical distribution's approximation ability, especially when the particle number is limited. In this paper, we propose to dynamically adjust particles' weights according to a Fisher-Rao reaction flow. We develop a general Dynamic-weight Particle-based Variational Inference (DPVI) framework according to a novel continuous composite flow, which evolves the positions and weights of particles simultaneously. We show that the mean-field limit of our composite flow is actually a Wasserstein-Fisher-Rao gradient flow of the associated dissimilarity functional. By using different finite-particle approximations in our general framework, we derive several efficient DPVI algorithms. The empirical results demonstrate the superiority of our derived DPVI algorithms over their fixed-weight counterparts.","doi":"10.24963/ijcai.2022/679","url":"https://ijcai.org/proceedings/2022/679","pdf":"https://ijcai.org/proceedings/2022/0679.pdf","markdown":"https://ijcai.org/proceedings/2022/679.md","bibtex":"https://ijcai.org/proceedings/2022/bibtex/679","volume":{"slug":"2022","title":"Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence","short_title":"IJCAI 2022","url":"https://ijcai.org/proceedings/2022/","year":2022,"location":"Vienna","dates":"23-29 July 2022","editors":"Luc De Raedt","isbn":"978-1-956792-00-3","publisher":"International Joint Conferences on Artificial Intelligence Organization","doi_prefix":"10.24963/ijcai.2022"},"section":"Uncertainty in AI","track":"Main Track","number":679,"pages":{"first":4900,"last":4906},"keywords":["Uncertainty in AI: Inference","Machine Learning: Bayesian Learning"],"videos":[{"label":"Video #1","url":"https://slideslive.com/38986648","length_sec":98,"provider":"slideslive"},{"label":"Video #2","url":"https://slideslive.com/38984791","length_sec":290,"provider":"slideslive"}],"references":[{"position":1,"raw":"[Ambrosio et al., 2008] Luigi Ambrosio, Nicola Gigli, and Giuseppe Savaré. Gradient flows: in metric spaces and in the space of probability measures. Springer Science & Business Media, 2008.","title":"Gradient flows: in metric spaces and in the space of probability measures","authors":"Ambrosio","year":"2008","venue":null,"doi":null,"url":null,"internal_url":null},{"position":2,"raw":"[Brooks et al., 2011] Steve Brooks, Andrew Gelman, Galin Jones, and Xiao-Li Meng. Handbook of markov chain monte carlo. CRC press, 2011.","title":"Handbook of markov chain monte carlo","authors":"Steve Brooks, Andrew Gelman, Galin Jones, Xiao-Li Meng","year":"2011","venue":null,"doi":null,"url":null,"internal_url":null},{"position":3,"raw":"[Butcher, 1964] John C Butcher. Implicit runge-kutta processes. Mathematics of Computation, 18(85):50-64, 1964.","title":"Implicit runge-kutta processes","authors":"C John, Butcher","year":"1964","venue":"Mathematics of Computation","doi":null,"url":null,"internal_url":null},{"position":4,"raw":"[Chen and Zhang, 2017] Changyou Chen and Ruiyi Zhang. Par- ticle optimization in stochastic gradient mcmc. arXiv preprint arXiv:1711.10927, 2017.","title":"Particle optimization in stochastic gradient mcmc","authors":"Chen, Zhang, Changyou Chen, Ruiyi Zhang","year":"2017","venue":null,"doi":null,"url":null,"internal_url":null},{"position":5,"raw":"[Chen et al., 2018a] Changyou Chen, Ruiyi Zhang, Wenlin Wang, Bai Li, and Liqun Chen. A unified particle-optimization framework for scalable bayesian sampling. arXiv preprint arXiv:1805.11659, 2018.","title":"A unified particle-optimization framework for scalable bayesian sampling","authors":"Chen","year":"2018","venue":null,"doi":null,"url":null,"internal_url":null},{"position":6,"raw":"[Chen et al., 2018b] Wilson Ye Chen, Lester Mackey, Jackson Gorham, Franc ¸ois-Xavier Briol, and Chris Oates. Stein points. In ICML, pages 844-853. PMLR, 2018.","title":"Stein points","authors":"Chen","year":"2018","venue":"ICML","doi":null,"url":null,"internal_url":null},{"position":7,"raw":"[Craig and Bertozzi, 2016] Katy Craig and Andrea Bertozzi. A blob method for the aggregation equation. Mathematics of com- putation, 85(300):1681-1717, 2016.","title":"A blob method for the aggregation equation","authors":"Craig, Bertozzi, Katy Craig, Andrea Bertozzi","year":"2016","venue":"Mathematics of computation","doi":null,"url":null,"internal_url":null},{"position":8,"raw":"[Gallouët and Monsaingeon, 2017] Thomas O Gallouët and Leonard Monsaingeon. A jko splitting scheme for kantorovich- fisher-rao gradient flows. SIAM Journal on Mathematical Analysis, 49(2):1100-1130, 2017.","title":"A jko splitting scheme for kantorovichfisher-rao gradient flows","authors":"O Thomas, Leonard Gallouët, Monsaingeon","year":"2017","venue":"SIAM Journal on Mathematical Analysis","doi":null,"url":null,"internal_url":null},{"position":9,"raw":"[Jordan et al., 1998] Richard Jordan, David Kinderlehrer, and Felix Otto. The variational formulation of the fokker-planck equation. SIAM journal on mathematical analysis, 29(1):1-17, 1998.","title":"The variational formulation of the fokker-planck equation","authors":"Richard Jordan, David Kinderlehrer, Felix Otto","year":"1998","venue":"SIAM journal on mathematical analysis","doi":null,"url":null,"internal_url":null},{"position":10,"raw":"[Kakutani, 1948] Shizuo Kakutani. On equivalence of infinite prod- uct measures. Annals of Mathematics, pages 214-224, 1948.","title":"On equivalence of infinite product measures","authors":"Shizuo Kakutani, Kakutani","year":"1948","venue":"Annals of Mathematics","doi":null,"url":null,"internal_url":null},{"position":11,"raw":"[Kondratyev et al., 2016] Stanislav Kondratyev, Léonard Monsain- geon, Dmitry Vorotnikov, et al. A new optimal transport distance on the space of finite radon measures. Advances in Differential Equations, 21(11/12):1117-1164, 2016.","title":"A new optimal transport distance on the space of finite radon measures","authors":"Stanislav Kondratyev, Léonard Monsaingeon, Dmitry Vorotnikov","year":"2016","venue":"Advances in Differential Equations","doi":null,"url":null,"internal_url":null},{"position":12,"raw":"[Korba et al., 2020] Anna Korba, Adil Salim, Michael Arbel, Giu- lia Luise, and Arthur Gretton. A non-asymptotic analysis for stein variational gradient descent. NeurIPS, 33, 2020.","title":"A non-asymptotic analysis for stein variational gradient descent","authors":"Anna Korba, Adil Salim, Michael Arbel, Giulia Luise, Arthur Gretton","year":"2020","venue":"NeurIPS","doi":null,"url":null,"internal_url":null},{"position":13,"raw":"[Korba et al., 2021] Anna Korba, Pierre-Cyril Aubin-Frankowski, Szymon Majewski, and Pierre Ablin. Kernel stein discrepancy descent. arXiv preprint arXiv:2105.09994, 2021.","title":"Kernel stein discrepancy descent","authors":"Anna Korba, Pierre-Cyril Aubin-Frankowski, Szymon Majewski, Pierre Ablin","year":"2021","venue":null,"doi":null,"url":null,"internal_url":null},{"position":14,"raw":"[Liero et al., 2016] Matthias Liero, Alexander Mielke, and Giuseppe Savaré. Optimal transport in competition with reac- tion: The hellinger-kantorovich distance and geodesic curves. SIAM Journal on Mathematical Analysis, 48(4):2869-2911, 2016.","title":"Optimal transport in competition with reaction: The hellinger-kantorovich distance and geodesic curves","authors":"Matthias Liero, Alexander Mielke, Giuseppe Savaré","year":"2016","venue":"SIAM Journal on Mathematical Analysis","doi":null,"url":null,"internal_url":null},{"position":15,"raw":"[Liu and Wang, 2016] Qiang Liu and Dilin Wang. Stein variational gradient descent: A general purpose bayesian inference algo- rithm. arXiv preprint arXiv:1608.04471, 2016.","title":"Stein variational gradient descent: A general purpose bayesian inference algorithm","authors":"Qiang Liu, Dilin Wang","year":"2016","venue":null,"doi":null,"url":null,"internal_url":null},{"position":16,"raw":"[Liu and Wang, 2018] Qiang Liu and Dilin Wang. Stein varia- tional gradient descent as moment matching. arXiv preprint arXiv:1810.11693, 2018.","title":"Stein variational gradient descent as moment matching","authors":"Qiang Liu, Dilin Wang","year":"2018","venue":null,"doi":null,"url":null,"internal_url":null},{"position":17,"raw":"[Liu and Zhu, 2018] Chang Liu and Jun Zhu. Riemannian stein variational gradient descent for bayesian inference. In AAAI, vol- ume 32, 2018.","title":"Riemannian stein variational gradient descent for bayesian inference","authors":"Chang Liu, Jun Zhu","year":"2018","venue":"AAAI","doi":null,"url":null,"internal_url":null},{"position":18,"raw":"[Liu et al., 2016] Qiang Liu, Jason Lee, and Michael Jordan. A kernelized stein discrepancy for goodness-of-fit tests. In ICML, pages 276-284. PMLR, 2016.","title":"A kernelized stein discrepancy for goodness-of-fit tests","authors":"Qiang Liu, Jason Lee, Michael Jordan","year":"2016","venue":"ICML","doi":null,"url":null,"internal_url":null},{"position":19,"raw":"[Liu et al., 2019a] Chang Liu, Jingwei Zhuo, Pengyu Cheng, Ruiyi Zhang, and Jun Zhu. Understanding and accelerating particle- based variational inference. In ICML, pages 4082-4092, 2019.","title":"Understanding and accelerating particlebased variational inference","authors":"Liu","year":"2019","venue":"ICML","doi":null,"url":null,"internal_url":null},{"position":20,"raw":"[Liu et al., 2019b] Chang Liu, Jingwei Zhuo, and Jun Zhu. Under- standing mcmc dynamics as flows on the wasserstein space. In ICML, 2019.","title":"Understanding mcmc dynamics as flows on the wasserstein space","authors":"Liu","year":"2019","venue":"ICML","doi":null,"url":null,"internal_url":null},{"position":21,"raw":"[Liu, 2017] Qiang Liu. Stein variational gradient descent as gradi- ent flow. arXiv preprint arXiv:1704.07520, 2017.","title":"Stein variational gradient descent as gradient flow","authors":"Qiang Liu","year":"2017","venue":null,"doi":null,"url":null,"internal_url":null},{"position":22,"raw":"[Lu et al., 2019] Yulong Lu, Jianfeng Lu, and James Nolen. Ac- celerating langevin sampling with birth-death. arXiv preprint arXiv:1905.09863, 2019.","title":"Accelerating langevin sampling with birth-death","authors":"Yulong Lu, Jianfeng Lu, James Nolen","year":"2019","venue":null,"doi":null,"url":null,"internal_url":null},{"position":23,"raw":"[Ma et al., 2015] Yi-An Ma, Tianqi Chen, and Emily B Fox. A complete recipe for stochastic gradient mcmc. arXiv preprint arXiv:1506.04696, 2015.","title":"A complete recipe for stochastic gradient mcmc","authors":"Yi-An Ma, Tianqi Chen, Emily B Fox","year":"2015","venue":null,"doi":null,"url":null,"internal_url":null},{"position":24,"raw":"[Mroueh and Rigotti, 2020] Youssef Mroueh and Mattia Rigotti. Unbalanced sobolev descent. NeurIPS, 33, 2020.","title":"Unbalanced sobolev descent","authors":"Youssef Mroueh, Mattia Rigotti","year":"2020","venue":"NeurIPS","doi":null,"url":null,"internal_url":null},{"position":25,"raw":"[Neal and others, 2011] Radford M Neal et al. Mcmc using hamil- tonian dynamics. Handbook of markov chain monte carlo, 2(11):2, 2011.","title":"Mcmc using hamiltonian dynamics","authors":"M Radford, Neal","year":"2011","venue":"Handbook of markov chain monte carlo","doi":null,"url":null,"internal_url":null},{"position":26,"raw":"[Platen and Bruti-Liberati, 2010] Eckhard Platen and Nicola Bruti- Liberati. Numerical solution of stochastic differential equations with jumps in finance, volume 64. Springer Science & Business Media, 2010.","title":"Numerical solution of stochastic differential equations with jumps in finance","authors":"Bruti-Liberati ; Platen, Eckhard Platen, Nicola Bruti-Liberati","year":"2010","venue":null,"doi":null,"url":null,"internal_url":null},{"position":27,"raw":"[Pu et al., 2017] Yunchen Pu, Zhe Gan, Ricardo Henao, Chunyuan Li, Shaobo Han, and Lawrence Carin. Vae learning via stein variational gradient descent. In NIPS, 2017.","title":"Vae learning via stein variational gradient descent","authors":"Yunchen Pu, Zhe Gan, Ricardo Henao, Chunyuan Li, Shaobo Han, Lawrence Carin","year":"2017","venue":"NIPS","doi":null,"url":null,"internal_url":null},{"position":28,"raw":"[Ranganath et al., 2014] Rajesh Ranganath, Sean Gerrish, and David Blei. Black box variational inference. In Artificial in- telligence and statistics, pages 814-822. PMLR, 2014.","title":"Black box variational inference","authors":"Rajesh Ranganath, Sean Gerrish, David Blei","year":"2014","venue":"Artificial intelligence and statistics","doi":null,"url":null,"internal_url":null},{"position":29,"raw":"[Rao, 1945] CR Rao. Information and accuracy attainable in the estimation of statistical parameters. kotz s & johnson nl (eds.), breakthroughs in statistics volume i: Foundations and basic the- ory, 235-248, 1945.","title":"Information and accuracy attainable in the estimation of statistical parameters","authors":"Rao","year":"1945","venue":"breakthroughs in statistics volume i: Foundations and basic theory","doi":null,"url":null,"internal_url":null},{"position":30,"raw":"[Rasmussen, 2003] Carl Edward Rasmussen. Gaussian processes in machine learning. In Summer school on machine learning, pages 63-71. Springer, 2003.","title":"Gaussian processes in machine learning","authors":"Carl Edward, Rasmussen","year":"2003","venue":"Summer school on machine learning","doi":null,"url":null,"internal_url":null},{"position":31,"raw":"[Rotskoff et al., 2019] Grant Rotskoff, Samy Jelassi, Joan Bruna, and Eric Vanden-Eijnden. Global convergence of neuron birth- death dynamics. arXiv preprint arXiv:1902.01843, 2019.","title":"Global convergence of neuron birthdeath dynamics","authors":"Grant Rotskoff, Samy Jelassi, Joan Bruna, Eric Vanden-Eijnden","year":"2019","venue":null,"doi":null,"url":null,"internal_url":null},{"position":32,"raw":"[Süli and Mayers, 2003] Endre Süli and David F Mayers. An intro- duction to numerical analysis. Cambridge university press, 2003.","title":"Endre Süli and David F Mayers. An introduction to numerical analysis","authors":"Mayers Süli","year":"2003","venue":null,"doi":null,"url":null,"internal_url":null},{"position":33,"raw":"[Wang and Li, 2019] Yifei Wang and Wuchen Li. Accelerated in- formation gradient flow. arXiv preprint arXiv:1909.02102, 2019.","title":"Accelerated information gradient flow","authors":"Wang, Li ;, Yifei Wang, Wuchen Li","year":"2019","venue":null,"doi":null,"url":null,"internal_url":null},{"position":34,"raw":"[Wibisono, 2018] Andre Wibisono. Sampling as optimization in the space of measures: The langevin dynamics as a composite optimization problem. In COLT, pages 2093-3027. PMLR, 2018.","title":"Sampling as optimization in the space of measures: The langevin dynamics as a composite optimization problem","authors":"Andre Wibisono","year":"2018","venue":"COLT","doi":null,"url":null,"internal_url":null},{"position":35,"raw":"[Zhang et al., 2020a] Jianyi Zhang, Ruiyi Zhang, Lawrence Carin, and Changyou Chen. Stochastic particle-optimization sampling and the non-asymptotic convergence theory. In Artificial Intelli- gence and Statistics, pages 1877-1887. PMLR, 2020.","title":"Stochastic particle-optimization sampling and the non-asymptotic convergence theory","authors":"Zhang","year":"2020","venue":"Artificial Intelligence and Statistics","doi":null,"url":null,"internal_url":null},{"position":36,"raw":"[Zhang et al., 2020b] Jianyi Zhang, Yang Zhao, and Changyou Chen. Variance reduction in stochastic particle-optimization sam- pling. In ICML, pages 11307-11316. PMLR, 2020.","title":"Variance reduction in stochastic particle-optimization sampling","authors":"Zhang","year":"2020","venue":"ICML","doi":null,"url":null,"internal_url":null},{"position":37,"raw":"[Zhu et al., 2020] Michael Zhu, Chang Liu, and Jun Zhu. Variance reduction and quasi-newton for particle-based variational infer- ence. In ICML, pages 11576-11587. PMLR, 2020.","title":"Variance reduction and quasi-newton for particle-based variational inference","authors":"Michael Zhu, Chang Liu, Jun Zhu","year":"2020","venue":"ICML","doi":null,"url":null,"internal_url":null},{"position":38,"raw":"[Zhuo et al., 2018] Jingwei Zhuo, Chang Liu, Jiaxin Shi, Jun Zhu, Ning Chen, and Bo Zhang. Message passing stein variational gradient descent. In ICML, pages 6018-6027. PMLR, 2018.","title":"Message passing stein variational gradient descent","authors":"Jingwei Zhuo, Chang Liu, Jiaxin Shi, Jun Zhu, Ning Chen, Bo Zhang","year":"2018","venue":"ICML","doi":null,"url":null,"internal_url":null}],"license_note":"Copyright International Joint Conferences on Artificial Intelligence."}