Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
116 tokens/sec
GPT-4o
10 tokens/sec
Gemini 2.5 Pro Pro
24 tokens/sec
o3 Pro
5 tokens/sec
GPT-4.1 Pro
3 tokens/sec
DeepSeek R1 via Azure Pro
35 tokens/sec
2000 character limit reached

Modern Bayesian Experimental Design (2302.14545v2)

Published 28 Feb 2023 in stat.ML, cs.AI, cs.LG, and stat.CO

Abstract: Bayesian experimental design (BED) provides a powerful and general framework for optimizing the design of experiments. However, its deployment often poses substantial computational challenges that can undermine its practical use. In this review, we outline how recent advances have transformed our ability to overcome these challenges and thus utilize BED effectively, before discussing some key areas for future development in the field.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (110)
  1. {barticle}[author] \bauthor\bsnmAtkinson, \bfnmAnthony C\binitsA. C. and \bauthor\bsnmFedorov, \bfnmValerii\binitsV. (\byear1975). \btitleThe design of experiments for discriminating between two rival models. \bjournalBiometrika. \endbibitem
  2. {barticle}[author] \bauthor\bsnmBarber, \bfnmDavid\binitsD. and \bauthor\bsnmAgakov, \bfnmFelix\binitsF. (\byear2003). \btitleThe IM algorithm: a variational approach to information maximization. \bjournalConference on Neural Information Processing Systems. \endbibitem
  3. {barticle}[author] \bauthor\bsnmBarnard, \bfnmGeorge A\binitsG. A., \bauthor\bsnmJenkins, \bfnmG M\binitsG. M. and \bauthor\bsnmWinsten, \bfnmCB\binitsC. (\byear1962). \btitleLikelihood inference and time series. \bjournalJournal of the Royal Statistical Society: Series A (General). \endbibitem
  4. {barticle}[author] \bauthor\bsnmBerger, \bfnmJames O\binitsJ. O. and \bauthor\bsnmWolpert, \bfnmRobert L\binitsR. L. (\byear1988). \btitleThe likelihood principle. \bjournalIMS Lecture Notes—Monograph Series. \endbibitem
  5. {barticle}[author] \bauthor\bsnmBernardo, \bfnmJosé M\binitsJ. M. (\byear1979). \btitleExpected information as expected utility. \bjournalThe Annals of Statistics. \endbibitem
  6. {barticle}[author] \bauthor\bsnmBhanot, \bfnmGyan\binitsG. and \bauthor\bsnmKennedy, \bfnmAnthony D\binitsA. D. (\byear1985). \btitleBosonic lattice gauge theory with noise. \bjournalPhysics Letters B. \endbibitem
  7. {barticle}[author] \bauthor\bsnmBlei, \bfnmDavid M\binitsD. M., \bauthor\bsnmKucukelbir, \bfnmAlp\binitsA. and \bauthor\bsnmMcAuliffe, \bfnmJon D\binitsJ. D. (\byear2017). \btitleVariational inference: a review for statisticians. \bjournalJournal of the American Statistical Association. \endbibitem
  8. {barticle}[author] \bauthor\bsnmBorth, \bfnmDavid M\binitsD. M. (\byear1975). \btitleA total entropy criterion for the dual problem of model discrimination and parameter estimation. \bjournalJournal of the Royal Statistical Society: Series B (Methodological). \endbibitem
  9. {btechreport}[author] \bauthor\bsnmBox, \bfnmGeorge EP\binitsG. E. (\byear1982). \btitleChoice of response surface design and alphabetic optimality. \btypeTechnical Report, \bpublisherUniversity of Wisconsin—Madison. \endbibitem
  10. {barticle}[author] \bauthor\bsnmBrynjarsdóttir, \bfnmJennỳ\binitsJ. and \bauthor\bsnmO’Hagan, \bfnmAnthony\binitsA. (\byear2014). \btitleLearning about physical parameters: the importance of model discrepancy. \bjournalInverse Problems. \endbibitem
  11. {barticle}[author] \bauthor\bsnmBurda, \bfnmYuri\binitsY., \bauthor\bsnmGrosse, \bfnmRoger\binitsR. and \bauthor\bsnmSalakhutdinov, \bfnmRuslan\binitsR. (\byear2015). \btitleImportance weighted autoencoders. \bjournalarXiv preprint arXiv:1509.00519. \endbibitem
  12. {barticle}[author] \bauthor\bsnmCarlin, \bfnmBradley P\binitsB. P., \bauthor\bsnmKadane, \bfnmJoseph B\binitsJ. B. and \bauthor\bsnmGelfand, \bfnmAlan E\binitsA. E. (\byear1998). \btitleApproaches for optimal sequential decision analysis in clinical trials. \bjournalBiometrics. \endbibitem
  13. {barticle}[author] \bauthor\bsnmChaloner, \bfnmKathryn\binitsK. and \bauthor\bsnmLarntz, \bfnmKinley\binitsK. (\byear1989). \btitleOptimal Bayesian design applied to logistic regression experiments. \bjournalJournal of Statistical Planning and Inference. \endbibitem
  14. {barticle}[author] \bauthor\bsnmChaloner, \bfnmKathryn\binitsK. and \bauthor\bsnmVerdinelli, \bfnmIsabella\binitsI. (\byear1995). \btitleBayesian experimental design: a review. \bjournalStatistical Science. \endbibitem
  15. {barticle}[author] \bauthor\bsnmCheng, \bfnmYi\binitsY. and \bauthor\bsnmShen, \bfnmYu\binitsY. (\byear2005). \btitleBayesian adaptive designs for clinical trials. \bjournalBiometrika. \endbibitem
  16. {btechreport}[author] \bauthor\bsnmClyde, \bfnmMerlise A\binitsM. A., \bauthor\bsnmMüller, \bfnmPeter\binitsP. and \bauthor\bsnmParmigiani, \bfnmGiovanni\binitsG. (\byear1996). \btitleExploring expected utility surfaces by Markov chains. \btypeTechnical Report. \endbibitem
  17. {barticle}[author] \bauthor\bsnmCook, \bfnmAlex R\binitsA. R., \bauthor\bsnmGibson, \bfnmGavin J\binitsG. J. and \bauthor\bsnmGilligan, \bfnmChristopher A\binitsC. A. (\byear2008). \btitleOptimal observation times in experimental epidemic processes. \bjournalBiometrics. \endbibitem
  18. {barticle}[author] \bauthor\bsnmCovey-Crump, \bfnmPAK\binitsP. and \bauthor\bsnmSilvey, \bfnmSD\binitsS. (\byear1970). \btitleOptimal regression designs with previous observations. \bjournalBiometrika. \endbibitem
  19. {barticle}[author] \bauthor\bsnmDehideniya, \bfnmMahasen B.\binitsM. B., \bauthor\bsnmDrovandi, \bfnmChristopher C.\binitsC. C. and \bauthor\bsnmMcGree, \bfnmJames M.\binitsJ. M. (\byear2018). \btitleOptimal Bayesian design for discriminating between models with intractable likelihoods in epidemiology. \bjournalComputational Statistics & Data Analysis. \endbibitem
  20. {barticle}[author] \bauthor\bsnmDrovandi, \bfnmChristopher C\binitsC. C. and \bauthor\bsnmPettitt, \bfnmAnthony N\binitsA. N. (\byear2013). \btitleBayesian experimental design for models with intractable likelihoods. \bjournalBiometrics. \endbibitem
  21. {barticle}[author] \bauthor\bsnmDushenko, \bfnmSergey\binitsS., \bauthor\bsnmAmbal, \bfnmKapildeb\binitsK. and \bauthor\bsnmMcMichael, \bfnmRobert D\binitsR. D. (\byear2020). \btitleSequential Bayesian experiment design for optically detected magnetic resonance of nitrogen-vacancy centers. \bjournalPhysical Review Applied. \endbibitem
  22. {barticle}[author] \bauthor\bsnmElfving, \bfnmGustav\binitsG. (\byear1952). \btitleOptimum allocation in linear regression theory. \bjournalThe Annals of Mathematical Statistics. \endbibitem
  23. {barticle}[author] \bauthor\bsnmEtzioni, \bfnmRuth\binitsR. and \bauthor\bsnmKadane, \bfnmJoseph B\binitsJ. B. (\byear1993). \btitleOptimal experimental design for another’s analysis. \bjournalJournal of the American Statistical Association. \endbibitem
  24. {barticle}[author] \bauthor\bsnmFarquhar, \bfnmSebastian\binitsS., \bauthor\bsnmGal, \bfnmYarin\binitsY. and \bauthor\bsnmRainforth, \bfnmTom\binitsT. (\byear2021). \btitleOn statistical bias in active learning: how and when to fix it. \bjournalInternational Conference on Learning Representations. \endbibitem
  25. {barticle}[author] \bauthor\bsnmFirth, \bfnmD\binitsD. and \bauthor\bsnmHinde, \bfnmJ P\binitsJ. P. (\byear1997). \btitleParameter neutral optimum design for non-linear models. \bjournalJournal of the Royal Statistical Society: Series B (Statistical Methodology). \endbibitem
  26. {barticle}[author] \bauthor\bsnmFisher, \bfnmRonald A\binitsR. A. (\byear1926). \btitleThe arrangement of field experiments. \bjournalJournal of the Ministry of Agriculture. \endbibitem
  27. {barticle}[author] \bauthor\bsnmFisher, \bfnmRonald A\binitsR. A. (\byear1936). \btitleThe design of experiments. \bjournalBritish Medical Journal. \endbibitem
  28. {bmisc}[author] \bauthor\bsnmUS Food and Drug Administration (\byear2019). \btitleAdaptive designs for clinical trials of drugs and biologics: guidance for industry. \endbibitem
  29. {barticle}[author] \bauthor\bsnmFudenberg, \bfnmDrew\binitsD., \bauthor\bsnmRomanyuk, \bfnmGleb\binitsG. and \bauthor\bsnmStrack, \bfnmPhilipp\binitsP. (\byear2017). \btitleActive learning with a misspecified prior. \bjournalTheoretical Economics. \endbibitem
  30. {barticle}[author] \bauthor\bsnmGal, \bfnmYarin\binitsY., \bauthor\bsnmIslam, \bfnmRiashat\binitsR. and \bauthor\bsnmGhahramani, \bfnmZoubin\binitsZ. (\byear2017). \btitleDeep Bayesian active learning with image data. \bjournalInternational Conference on Machine Learning. \endbibitem
  31. {barticle}[author] \bauthor\bsnmGelman, \bfnmAndrew\binitsA. (\byear2008). \btitleObjections to Bayesian statistics. \bjournalBayesian Analysis. \endbibitem
  32. {barticle}[author] \bauthor\bsnmGiles, \bfnmMichael B\binitsM. B. (\byear2008). \btitleMultilevel Monte Carlo path simulation. \bjournalOperations Research. \endbibitem
  33. {barticle}[author] \bauthor\bsnmGiles, \bfnmMichael B\binitsM. B. and \bauthor\bsnmSzpruch, \bfnmLukasz\binitsL. (\byear2014). \btitleAntithetic multilevel Monte Carlo estimation for multi-dimensional SDEs without Lévy area simulation. \bjournalThe Annals of Applied Probability. \endbibitem
  34. {barticle}[author] \bauthor\bsnmGillespie, \bfnmColin S\binitsC. S. and \bauthor\bsnmBoys, \bfnmRichard J\binitsR. J. (\byear2019). \btitleEfficient construction of Bayes optimal designs for stochastic process models. \bjournalStatistics and Computing. \endbibitem
  35. {barticle}[author] \bauthor\bsnmGo, \bfnmJinwoo\binitsJ. and \bauthor\bsnmIsaac, \bfnmTobin\binitsT. (\byear2022). \btitleRobust expected information gain for optimal Bayesian experimental design using ambiguity sets. \bjournalConference on Uncertainty in Artificial Intelligence. \endbibitem
  36. {barticle}[author] \bauthor\bsnmGoda, \bfnmTakashi\binitsT., \bauthor\bsnmHironaka, \bfnmTomohiko\binitsT. and \bauthor\bsnmIwamoto, \bfnmTakeru\binitsT. (\byear2020). \btitleMultilevel Monte Carlo estimation of expected information gains. \bjournalStochastic Analysis and Applications. \endbibitem
  37. {barticle}[author] \bauthor\bsnmGolovin, \bfnmDaniel\binitsD., \bauthor\bsnmKrause, \bfnmAndreas\binitsA. and \bauthor\bsnmRay, \bfnmDebajyoti\binitsD. (\byear2010). \btitleNear-optimal Bayesian active learning with noisy observations. \bjournalConference on Neural Information Processing Systems. \endbibitem
  38. {barticle}[author] \bauthor\bsnmGonzález, \bfnmJavier\binitsJ., \bauthor\bsnmOsborne, \bfnmMichael\binitsM. and \bauthor\bsnmLawrence, \bfnmNeil\binitsN. (\byear2016). \btitleGLASSES: relieving the myopia of Bayesian optimisation. \bjournalInternational Conference on Artificial Intelligence and Statistics. \endbibitem
  39. {barticle}[author] \bauthor\bsnmGrünwald, \bfnmPeter\binitsP. and \bauthor\bparticlevan \bsnmOmmen, \bfnmThijs\binitsT. (\byear2017). \btitleInconsistency of Bayesian inference for misspecified linear models, and a proposal for repairing it. \bjournalBayesian Analysis. \endbibitem
  40. {barticle}[author] \bauthor\bsnmGuez, \bfnmArthur\binitsA., \bauthor\bsnmSilver, \bfnmDavid\binitsD. and \bauthor\bsnmDayan, \bfnmPeter\binitsP. (\byear2012). \btitleEfficient Bayes-adaptive reinforcement learning using sample-based search. \bjournalConference on Neural Information Processing Systems. \endbibitem
  41. {barticle}[author] \bauthor\bsnmHainy, \bfnmMarkus\binitsM., \bauthor\bsnmDrovandi, \bfnmChristopher C.\binitsC. C. and \bauthor\bsnmMcGree, \bfnmJames M.\binitsJ. M. (\byear2016). \btitleLikelihood-free extensions for Bayesian sequentially designed experiments. \bjournalmODa 11—Advances in Model-Oriented Design and Analysis. \endbibitem
  42. {barticle}[author] \bauthor\bsnmHan, \bfnmCong\binitsC. and \bauthor\bsnmChaloner, \bfnmKathryn\binitsK. (\byear2004). \btitleBayesian experimental design for nonlinear mixed-effects models with application to HIV dynamics. \bjournalBiometrics. \endbibitem
  43. {barticle}[author] \bauthor\bsnmHennig, \bfnmPhilipp\binitsP. and \bauthor\bsnmSchuler, \bfnmChristian J\binitsC. J. (\byear2012). \btitleEntropy search for information-efficient global optimization. \bjournalJournal of Machine Learning Research. \endbibitem
  44. {barticle}[author] \bauthor\bsnmHuan, \bfnmXun\binitsX. and \bauthor\bsnmMarzouk, \bfnmYoussef\binitsY. (\byear2014). \btitleGradient-based stochastic optimization methods in Bayesian experimental design. \bjournalInternational Journal for Uncertainty Quantification. \endbibitem
  45. {barticle}[author] \bauthor\bsnmHuan, \bfnmXun\binitsX. and \bauthor\bsnmMarzouk, \bfnmYoussef M\binitsY. M. (\byear2013). \btitleSimulation-based optimal Bayesian experimental design for nonlinear systems. \bjournalJournal of Computational Physics. \endbibitem
  46. {barticle}[author] \bauthor\bsnmHuan, \bfnmXun\binitsX. and \bauthor\bsnmMarzouk, \bfnmYoussef M\binitsY. M. (\byear2016). \btitleSequential Bayesian optimal experimental design via approximate dynamic programming. \bjournalarXiv preprint arXiv:1604.08320. \endbibitem
  47. {barticle}[author] \bauthor\bsnmHughes-Oliver, \bfnmJacqueline M\binitsJ. M. (\byear1998). \btitleOptimal designs for nonlinear models with correlated errors. \bjournalIMS Lecture Notes—Monograph Series. \endbibitem
  48. {barticle}[author] \bauthor\bsnmJörke, \bfnmMatthew\binitsM., \bauthor\bsnmLee, \bfnmJonathan\binitsJ. and \bauthor\bsnmBrunskill, \bfnmEmma\binitsE. (\byear2022). \btitleSimple regret minimization for contextual bandits using Bayesian optimal experimental design. \bjournalICML Workshop on “Adaptive Experimental Design and Active Learning in the Real World”. \endbibitem
  49. {barticle}[author] \bauthor\bsnmKasy, \bfnmMaximilian\binitsM. and \bauthor\bsnmSautmann, \bfnmAnja\binitsA. (\byear2021). \btitleAdaptive treatment assignment in experiments for policy choice. \bjournalEconometrica. \endbibitem
  50. {barticle}[author] \bauthor\bsnmKennamer, \bfnmNoble\binitsN., \bauthor\bsnmWalton, \bfnmSteven\binitsS. and \bauthor\bsnmIhler, \bfnmAlexander\binitsA. (\byear2023). \btitleDesign amortization for Bayesian optimal experimental design. \bjournalAAAI Conference on Artificial Intelligence. \endbibitem
  51. {barticle}[author] \bauthor\bsnmKennedy, \bfnmAnthony D\binitsA. D. and \bauthor\bsnmKuti, \bfnmJulius\binitsJ. (\byear1985). \btitleNoise without noise: a new Monte Carlo method. \bjournalPhysical Review Letters. \endbibitem
  52. {barticle}[author] \bauthor\bsnmKennedy, \bfnmMarc C\binitsM. C. and \bauthor\bsnmO’Hagan, \bfnmAnthony\binitsA. (\byear2001). \btitleBayesian calibration of computer models. \bjournalJournal of the Royal Statistical Society: Series B (Statistical Methodology). \endbibitem
  53. {barticle}[author] \bauthor\bsnmKingma, \bfnmDiederik P\binitsD. P. and \bauthor\bsnmBa, \bfnmJimmy\binitsJ. (\byear2015). \btitleAdam: a method for stochastic optimization. \bjournalInternational Conference on Learning Representations. \endbibitem
  54. {barticle}[author] \bauthor\bsnmKingma, \bfnmDiederik P\binitsD. P. and \bauthor\bsnmWelling, \bfnmMax\binitsM. (\byear2014). \btitleAuto-encoding variational Bayes. \bjournalInternational Conference on Learning Representations. \endbibitem
  55. {barticle}[author] \bauthor\bsnmKirsch, \bfnmAndreas\binitsA., \bauthor\bsnmvan Amersfoort, \bfnmJoost\binitsJ. and \bauthor\bsnmGal, \bfnmYarin\binitsY. (\byear2019). \btitleBatchBALD: efficient and diverse batch acquisition for deep Bayesian active learning. \bjournalConference on Neural Information Processing Systems. \endbibitem
  56. {barticle}[author] \bauthor\bsnmKleijn, \bfnmBas JK\binitsB. J. and \bauthor\bparticlevan der \bsnmVaart, \bfnmAad W\binitsA. W. (\byear2012). \btitleThe Bernstein-von-Mises theorem under misspecification. \bjournalElectronic Journal of Statistics. \endbibitem
  57. {barticle}[author] \bauthor\bsnmKleinegesse, \bfnmS.\binitsS. and \bauthor\bsnmGutmann, \bfnmM. U.\binitsM. U. (\byear2019). \btitleEfficient Bayesian experimental design for implicit models. \bjournalInternational Conference on Artificial Intelligence and Statistics. \endbibitem
  58. {barticle}[author] \bauthor\bsnmKleinegesse, \bfnmSteven\binitsS. and \bauthor\bsnmGutmann, \bfnmMichael\binitsM. (\byear2020). \btitleBayesian experimental design for implicit models by mutual information neural estimation. \bjournalInternational Conference on Machine Learning. \endbibitem
  59. {barticle}[author] \bauthor\bsnmKleinegesse, \bfnmSteven\binitsS. and \bauthor\bsnmGutmann, \bfnmMichael U.\binitsM. U. (\byear2021). \btitleGradient-based Bayesian experimental design for implicit models using mutual information lower bounds. \bjournalarXiv preprint arXiv:2105.04379. \endbibitem
  60. {barticle}[author] \bauthor\bsnmKuck, \bfnmHendrik\binitsH., \bauthor\bparticlede \bsnmFreitas, \bfnmNando\binitsN. and \bauthor\bsnmDoucet, \bfnmArnaud\binitsA. (\byear2006). \btitleSMC samplers for Bayesian optimal nonlinear design. \bjournalIEEE Nonlinear Statistical Signal Processing Workshop. \endbibitem
  61. {barticle}[author] \bauthor\bsnmKuhfeld, \bfnmWarren F\binitsW. F., \bauthor\bsnmTobias, \bfnmRandall D\binitsR. D. and \bauthor\bsnmGarratt, \bfnmMark\binitsM. (\byear1994). \btitleEfficient experimental design with marketing research applications. \bjournalJournal of Marketing Research. \endbibitem
  62. {barticle}[author] \bauthor\bsnmLewi, \bfnmJeremy\binitsJ., \bauthor\bsnmButera, \bfnmRobert\binitsR. and \bauthor\bsnmPaninski, \bfnmLiam\binitsL. (\byear2009). \btitleSequential optimal design of neurophysiology experiments. \bjournalNeural Computation. \endbibitem
  63. {barticle}[author] \bauthor\bsnmLindley, \bfnmDennis V\binitsD. V. (\byear1956). \btitleOn a measure of the information provided by an experiment. \bjournalThe Annals of Mathematical Statistics. \endbibitem
  64. {bbook}[author] \bauthor\bsnmLindley, \bfnmDennis V\binitsD. V. (\byear1972). \btitleBayesian Statistics, a Review. \bpublisherSociety for Industrial and Applied Mathematics. \endbibitem
  65. {barticle}[author] \bauthor\bsnmLong, \bfnmQuan\binitsQ. (\byear2022). \btitleMultimodal information gain in Bayesian design of experiments. \bjournalComputational Statistics. \endbibitem
  66. {barticle}[author] \bauthor\bsnmLoredo, \bfnmThomas J\binitsT. J. (\byear2004). \btitleBayesian adaptive exploration. \bjournalAIP Conference Proceedings. \endbibitem
  67. {barticle}[author] \bauthor\bsnmMacKay, \bfnmDavid JC\binitsD. J. (\byear1992). \btitleInformation-based objective functions for active data selection. \bjournalNeural Computation. \endbibitem
  68. {barticle}[author] \bauthor\bsnmMargatina, \bfnmKaterina\binitsK., \bauthor\bsnmBarrault, \bfnmLoïc\binitsL. and \bauthor\bsnmAletras, \bfnmNikolaos\binitsN. (\byear2021). \btitleBayesian active learning with pretrained language models. \bjournalarXiv preprint arXiv:2104.08320. \endbibitem
  69. {barticle}[author] \bauthor\bsnmMcLeish, \bfnmDon\binitsD. (\byear2011). \btitleA general method for debiasing a Monte Carlo estimator. \bjournalMonte Carlo Methods and Applications. \endbibitem
  70. {barticle}[author] \bauthor\bsnmMeyer, \bfnmRuth K\binitsR. K. and \bauthor\bsnmNachtsheim, \bfnmChristopher J\binitsC. J. (\byear1995). \btitleThe coordinate-exchange algorithm for constructing exact optimal experimental designs. \bjournalTechnometrics. \endbibitem
  71. {barticle}[author] \bauthor\bsnmMüller, \bfnmPeter\binitsP. (\byear2005). \btitleSimulation based optimal design. \bjournalHandbook of Statistics. \endbibitem
  72. {barticle}[author] \bauthor\bsnmMüller, \bfnmPeter\binitsP., \bauthor\bsnmSansó, \bfnmBruno\binitsB. and \bauthor\bsnmDe Iorio, \bfnmMaria\binitsM. (\byear2004). \btitleOptimal Bayesian design by inhomogeneous Markov chain simulation. \bjournalJournal of the American Statistical Association. \endbibitem
  73. {barticle}[author] \bauthor\bsnmMyung, \bfnmJay I\binitsJ. I., \bauthor\bsnmCavagnaro, \bfnmDaniel R\binitsD. R. and \bauthor\bsnmPitt, \bfnmMark A\binitsM. A. (\byear2013). \btitleA tutorial on adaptive design optimization. \bjournalJournal of Mathematical Psychology. \endbibitem
  74. {barticle}[author] \bauthor\bsnmNeiswanger, \bfnmWillie\binitsW., \bauthor\bsnmWang, \bfnmKe Alexander\binitsK. A. and \bauthor\bsnmErmon, \bfnmStefano\binitsS. (\byear2021). \btitleBayesian algorithm execution: estimating computable properties of black-box functions using mutual information. \bjournalInternational Conference on Machine Learning. \endbibitem
  75. {barticle}[author] \bauthor\bsnmNguyen, \bfnmXuanlong\binitsX., \bauthor\bsnmWainwright, \bfnmMartin J.\binitsM. J. and \bauthor\bsnmJordan, \bfnmMichael I.\binitsM. I. (\byear2010). \btitleEstimating divergence functionals and the likelihood ratio by convex risk minimization. \bjournalIEEE Transactions on Information Theory. \endbibitem
  76. {barticle}[author] \bauthor\bsnmOverstall, \bfnmAntony\binitsA. and \bauthor\bsnmMcGree, \bfnmJames\binitsJ. (\byear2020). \btitleBayesian design of experiments for intractable likelihood models using coupled auxiliary models and multivariate emulation. \bjournalBayesian Analysis. \endbibitem
  77. {barticle}[author] \bauthor\bsnmOverstall, \bfnmAntony\binitsA. and \bauthor\bsnmMcGree, \bfnmJames\binitsJ. (\byear2021). \btitleBayesian decision-theoretic design of experiments under an alternative model. \bjournalBayesian Analysis. \endbibitem
  78. {barticle}[author] \bauthor\bsnmOverstall, \bfnmAntony M\binitsA. M. (\byear2022). \btitleProperties of Fisher information gain for Bayesian design of experiments. \bjournalJournal of Statistical Planning and Inference. \endbibitem
  79. {barticle}[author] \bauthor\bsnmOverstall, \bfnmAntony M\binitsA. M. and \bauthor\bsnmWoods, \bfnmDavid C\binitsD. C. (\byear2017). \btitleBayesian design of experiments using approximate coordinate exchange. \bjournalTechnometrics. \endbibitem
  80. {barticle}[author] \bauthor\bsnmPacheco, \bfnmJason\binitsJ. and \bauthor\bsnmFisher, \bfnmJohn\binitsJ. (\byear2019). \btitleVariational information planning for sequential decision making. \bjournalInternational Conference on Artificial Intelligence and Statistics. \endbibitem
  81. {barticle}[author] \bauthor\bsnmPaninski, \bfnmLiam\binitsL. (\byear2005). \btitleAsymptotic theory of information-theoretic experimental design. \bjournalNeural Computation. \endbibitem
  82. {barticle}[author] \bauthor\bsnmPapadimitriou, \bfnmCostas\binitsC. (\byear2004). \btitleOptimal sensor placement methodology for parametric identification of structural systems. \bjournalJournal of Sound and Vibration. \endbibitem
  83. {barticle}[author] \bauthor\bsnmPrangle, \bfnmDennis\binitsD., \bauthor\bsnmHarbisher, \bfnmSophie\binitsS. and \bauthor\bsnmGillespie, \bfnmColin S\binitsC. S. (\byear2022). \btitleBayesian experimental design without posterior calculations: an adversarial approach. \bjournalBayesian Analysis. \endbibitem
  84. {barticle}[author] \bauthor\bsnmRezende, \bfnmDanilo Jimenez\binitsD. J., \bauthor\bsnmMohamed, \bfnmShakir\binitsS. and \bauthor\bsnmWierstra, \bfnmDaan\binitsD. (\byear2014). \btitleStochastic backpropagation and approximate inference in deep generative models. \bjournalInternational Conference on Machine Learning. \endbibitem
  85. {barticle}[author] \bauthor\bsnmRhee, \bfnmChang-han\binitsC.-h. and \bauthor\bsnmGlynn, \bfnmPeter W\binitsP. W. (\byear2015). \btitleUnbiased estimation with square root convergence for SDE models. \bjournalOperations Research. \endbibitem
  86. {barticle}[author] \bauthor\bsnmRobbins, \bfnmHerbert\binitsH. and \bauthor\bsnmMonro, \bfnmSutton\binitsS. (\byear1951). \btitleA stochastic approximation method. \bjournalThe Annals of Mathematical Statistics. \endbibitem
  87. {barticle}[author] \bauthor\bsnmRoss, \bfnmStephane\binitsS., \bauthor\bsnmChaib-draa, \bfnmBrahim\binitsB. and \bauthor\bsnmPineau, \bfnmJoelle\binitsJ. (\byear2007). \btitleBayes-adaptive POMDPs. \bjournalConference on Neural Information Processing Systems. \endbibitem
  88. {barticle}[author] \bauthor\bsnmRyan, \bfnmElizabeth G\binitsE. G., \bauthor\bsnmDrovandi, \bfnmChristopher C\binitsC. C. and \bauthor\bsnmPettitt, \bfnmAnthony N\binitsA. N. (\byear2015). \btitleFully Bayesian experimental design for pharmacokinetic studies. \bjournalEntropy. \endbibitem
  89. {barticle}[author] \bauthor\bsnmRyan, \bfnmKenneth J\binitsK. J. (\byear2003). \btitleEstimating expected information gains for experimental designs with application to the random fatigue-limit model. \bjournalJournal of Computational and Graphical Statistics. \endbibitem
  90. {barticle}[author] \bauthor\bsnmRyan, \bfnmThomas P\binitsT. P. and \bauthor\bsnmMorgan, \bfnmJP\binitsJ. (\byear2007). \btitleModern experimental design. \bjournalJournal of Statistical Theory and Practice. \endbibitem
  91. {barticle}[author] \bauthor\bsnmSebastiani, \bfnmPaola\binitsP. and \bauthor\bsnmWynn, \bfnmHenry P\binitsH. P. (\byear2000). \btitleMaximum entropy sampling and optimal Bayesian experimental design. \bjournalJournal of the Royal Statistical Society: Series B (Statistical Methodology). \endbibitem
  92. {btechreport}[author] \bauthor\bsnmSettles, \bfnmBurr\binitsB. (\byear2009). \btitleActive learning literature survey. \btypeTechnical Report, \bpublisherUniversity of Wisconsin—Madison. \endbibitem
  93. {barticle}[author] \bauthor\bsnmShannon, \bfnmClaude Elwood\binitsC. E. (\byear1948). \btitleA mathematical theory of communication. \bjournalThe Bell System Technical Journal. \endbibitem
  94. {barticle}[author] \bauthor\bsnmShen, \bfnmWanggang\binitsW. and \bauthor\bsnmHuan, \bfnmXun\binitsX. (\byear2023). \btitleBayesian sequential optimal experimental design for nonlinear models using policy gradient reinforcement learning. \bjournalComputer Methods in Applied Mechanics and Engineering. \endbibitem
  95. {barticle}[author] \bauthor\bsnmSiddhant, \bfnmAditya\binitsA. and \bauthor\bsnmLipton, \bfnmZachary C\binitsZ. C. (\byear2018). \btitleDeep Bayesian active learning for natural language processing: results of a large-scale empirical study. \bjournalarXiv preprint arXiv:1808.05697. \endbibitem
  96. {barticle}[author] \bauthor\bsnmSmith, \bfnmKirstine\binitsK. (\byear1918). \btitleOn the standard deviations of adjusted and interpolated values of an observed polynomial function and its constants and the guidance they give towards a proper choice of the distribution of observations. \bjournalBiometrika. \endbibitem
  97. {barticle}[author] \bauthor\bsnmStroud, \bfnmJonathan R\binitsJ. R., \bauthor\bsnmMüller, \bfnmPeter\binitsP. and \bauthor\bsnmRosner, \bfnmGary L\binitsG. L. (\byear2001). \btitleOptimal sampling times in population pharmacokinetic studies. \bjournalJournal of the Royal Statistical Society: Series C (Applied Statistics). \endbibitem
  98. {barticle}[author] \bauthor\bsnmStuhlmüller, \bfnmAndreas\binitsA., \bauthor\bsnmTaylor, \bfnmJacob\binitsJ. and \bauthor\bsnmGoodman, \bfnmNoah\binitsN. (\byear2013). \btitleLearning stochastic inverses. \bjournalConference on Neural Information Processing Systems. \endbibitem
  99. {barticle}[author] \bauthor\bsnmSun, \bfnmYi\binitsY., \bauthor\bsnmGomez, \bfnmFaustino\binitsF. and \bauthor\bsnmSchmidhuber, \bfnmJürgen\binitsJ. (\byear2011). \btitlePlanning to be surprised: optimal Bayesian exploration in dynamic environments. \bjournalInternational Conference on Artificial General Intelligence. \endbibitem
  100. {bbook}[author] \bauthor\bsnmSutton, \bfnmRichard S\binitsR. S. and \bauthor\bsnmBarto, \bfnmAndrew G\binitsA. G. (\byear2018). \btitleReinforcement Learning: an Introduction. \bpublisherMIT Press. \endbibitem
  101. {barticle}[author] \bauthor\bsnmSzepesvári, \bfnmCsaba\binitsC. (\byear2010). \btitleAlgorithms for Reinforcement Learning. \bjournalSynthesis Lectures on Artificial Intelligence and Machine Learning. \endbibitem
  102. {barticle}[author] \bauthor\bsnmTsutakawa, \bfnmRobert K\binitsR. K. (\byear1972). \btitleDesign of experiment for bioassay. \bjournalJournal of the American Statistical Association. \endbibitem
  103. {barticle}[author] \bauthor\bparticlevan \bsnmDen Berg, \bfnmJojanneke\binitsJ., \bauthor\bsnmCurtis, \bfnmAndrew\binitsA. and \bauthor\bsnmTrampert, \bfnmJeannot\binitsJ. (\byear2003). \btitleOptimal nonlinear Bayesian experimental design: an application to amplitude versus offset experiments. \bjournalGeophysical Journal International. \endbibitem
  104. {barticle}[author] \bauthor\bparticlevan den \bsnmOord, \bfnmAäron\binitsA., \bauthor\bsnmLi, \bfnmYazhe\binitsY. and \bauthor\bsnmVinyals, \bfnmOriol\binitsO. (\byear2018). \btitleRepresentation learning with contrastive predictive coding. \bjournalarXiv preprint arXiv:1807.03748. \endbibitem
  105. {barticle}[author] \bauthor\bparticlevon \bsnmKügelgen, \bfnmJulius\binitsJ., \bauthor\bsnmRubenstein, \bfnmPaul K\binitsP. K., \bauthor\bsnmSchölkopf, \bfnmBernhard\binitsB. and \bauthor\bsnmWeller, \bfnmAdrian\binitsA. (\byear2019). \btitleOptimal experimental design via Bayesian optimization: active causal structure learning for Gaussian process networks. \bjournalNeurIPS Workshop on “Do the Right Thing: Machine Learning and Causal Inference for Improved Decision Making”. \endbibitem
  106. {barticle}[author] \bauthor\bsnmWalker, \bfnmStephen G\binitsS. G. (\byear2016). \btitleBayesian information in an experiment and the Fisher information distance. \bjournalStatistics & Probability Letters. \endbibitem
  107. {barticle}[author] \bauthor\bsnmWang, \bfnmZi\binitsZ. and \bauthor\bsnmJegelka, \bfnmStefanie\binitsS. (\byear2017). \btitleMax-value entropy search for efficient Bayesian optimization. \bjournalInternational Conference on Machine Learning. \endbibitem
  108. {barticle}[author] \bauthor\bsnmWatson, \bfnmAndrew B\binitsA. B. (\byear2017). \btitleQUEST+: a general multidimensional Bayesian adaptive psychometric method. \bjournalJournal of Vision. \endbibitem
  109. {barticle}[author] \bauthor\bsnmZhang, \bfnmJiaxin\binitsJ., \bauthor\bsnmBi, \bfnmSirui\binitsS. and \bauthor\bsnmZhang, \bfnmGuannan\binitsG. (\byear2021). \btitleA scalable gradient-free method for Bayesian experimental design with implicit models. \bjournalInternational Conference on Artificial Intelligence and Statistics. \endbibitem
  110. {barticle}[author] \bauthor\bsnmZheng, \bfnmSue\binitsS., \bauthor\bsnmPacheco, \bfnmJason\binitsJ. and \bauthor\bsnmFisher, \bfnmJohn\binitsJ. (\byear2018). \btitleA robust approach to sequential information theoretic planning. \bjournalInternational Conference on Machine Learning. \endbibitem
Citations (51)

Summary

  • The paper demonstrates that recent innovations in Bayesian experimental design overcome computational challenges and enable scalable optimization.
  • It highlights the effectiveness of debiasing strategies, variational methods, and policy-based approaches in maximizing Expected Information Gain.
  • The advancements pave the way for broader applications in clinical trials, physics simulations, and market research, urging further research exploration.

Bayesian Experimental Design: Advancements and Future Directions

The paper "Modern Bayesian Experimental Design" by Tom Rainforth et al. provides a comprehensive overview of recent advancements in Bayesian Experimental Design (BED) and its adaptive counterpart, BAD. The authors illuminate the transformative potential of BED in optimizing experimental setups across a diverse array of fields including clinical trials, market research, and physics simulations, among others. Nonetheless, despite its theoretical strengths, BED has faced significant computational challenges that have historically impeded its wider adoption. This paper elucidates how recent methodological developments have mitigated many of these obstacles, thus paving the way for more extensive and effective deployment of BED methodologies.

Bayesian Experimental Design and Optimization

BED revolves around the optimization of experimental designs through information-theoretic concepts, specifically by maximizing the Expected Information Gain (EIG). The EIG quantifies the anticipated gains in information about the parameters of interest θ\theta from the experimental outcomes. In adaptive settings, BAD iteratively refines designs by updating them based on information gathered in prior steps, thereby embodying a sequential decision-making process. However, traditional BAD approaches have been critiqued for their computational inefficiencies and inherent myopic nature since they often do not account for the long-term implications of intermediate design decisions.

Computational Challenges and Solutions

Historically, the computational infeasibility of BED arises from the intricacies of nested estimation required to compute the EIG. Methods such as Nested Monte Carlo (NMC) estimators have been employed, but these typically involve high computational cost due to their inherent bias and slow convergence rates. Recent breakthroughs have seen the emergence of debiasing strategies and novel estimation techniques that provide unbiased gradient estimates, which are crucial for effective and efficient optimization.

Functional and variational approaches have garnered attention due to their flexibility and reduced computational burden. Variational methods, in particular, leverage surrogate functions to approximate intractable terms, thereby enabling consistent and scalable EIG estimations. The application of these methodologies marks a shift towards more practical and agile implementations of BED, extending its applicability to high-dimensional, continuous, as well as implicit models.

Policy-Based Approaches

A pivotal advancement discussed is the shift towards policy-based approaches to BAD. Traditional BAD has been inherently sequential and computationally intensive. The introduction of a policy network concept, exemplified by the Deep Adaptive Design (DAD) framework, significantly alleviates these constraints by offloading complex computation to an offline training phase. Policies pre-trained to map previous experimental data to design decisions eliminate the need for on-the-fly computation, allowing near-instantaneous adaptation during live experiments. Such strategies also foster non-myopic design decisions, improving the overall efficacy and efficiency of the experimental designs being implemented.

Implications and Future Research

The advances reviewed in this paper hold substantial implications for the future of BED. They challenge the limitations imposed by traditional methodologies and open up new avenues for research. Opportunities exist in scaling these methods to more complex and higher-dimensional problems, improving resilience to model misspecification, and better integrating BED with related fields such as Bayesian active learning and reinforcement learning. There is also potential in refining the interplay between experiment design and downstream analysis, especially concerning robustness and informativeness of collected data.

In conclusion, the paper provides a vivid depiction of how BED has evolved with recent computational advancements, steering it toward broader applicability and effectiveness. As BED continues to integrate cutting-edge computational techniques and leverage interdisciplinary insights, its impact across scientific and industrial domains is poised to expand significantly. Future research will be crucial in consolidating these gains and ensuring that BED realizes its full potential in an array of experimental contexts.

Dice Question Streamline Icon: https://streamlinehq.com

Follow-up Questions

We haven't generated follow-up questions for this paper yet.

X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets

Youtube Logo Streamline Icon: https://streamlinehq.com