Mining gold from implicit models to improve likelihood-free inference
Open accessThe DOI matches its Crossref record; 64 of the 64 references it cites were found in a registry or an index.
Abstract
Simulators often provide the best description of real-world phenomena. However, the probability density that they implicitly define is often intractable, leading to challenging inverse problems for inference. Recently, a number of techniques have been introduced in which a surrogate for the intractable density is learned, including normalizing flows and density ratio estimators. We show that additional information that characterizes the latent process can often be extracted from simulators and used to augment the training data for these surrogate models. We introduce several loss functions that leverage these augmented data and demonstrate that these techniques can improve sample efficiency and quality of inference.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Advances in Neural Information Processing Systems 29
Teymur, O., Zygalakis, K., & Calderhead, B. (2016) · Cited by 13,887
Alwall, J., Frederix, R., Frixione, S., Hirschi, V., Maltoni, F., Mattelaer, O., Shao, H.-S., Stelzer, T., Torrielli, P., & Zaro, M. (2014) · Journal of High Energy Physics · doi:10.1007/jhep07(2014)079 · Cited by 7,579 · cited 1×
“As simulator we use a combination of MadGraph 5 [43] and MadMax [44–46].”
MethodGillespie, D. T. (1976) · Journal of Computational Physics · doi:10.1016/0021-9991(76)90041-3 · Cited by 6,029 · cited 1×
“We simulate the Lotka-Volterra model with Gillespie’s algorithm [50].”
MethodWaveNet: a generative model for raw audio
Oord, A. V. D., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A., Kalchbrenner, N., Senior, A., & Kavukcuoglu, K. (2016) · arXiv · doi:10.48550/arxiv.1609.03499 · Cited by 4,911
Generative Adversarial Networks
Goodfellow, I. J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1406.2661 · Cited by 4,477 · cited 1×
“This “likelihood ratio trick” or “density ratio trick” is widely appreciated [9, 20, 26, 31].”
The Large-Sample Distribution of the Likelihood Ratio for Testing Composite Hypotheses
Wilks, S. S. (1938) · The Annals of Mathematical Statistics · doi:10.1214/aoms/1177732360 · Cited by 3,493
Simple Statistical Gradient-Following Algorithms for Connectionist Reinforcement Learning
Williams, R. J. (1992) · Machine Learning · doi:10.1007/bf00992696 · Cited by 3,416 · cited 1×
“This is essentially the same observation as the policy gradient used in REINFORCE [23]; however, instead of trying to optimize θ via a stochastic gradient estimate of some reward function, we will simply augment the data generated by the simulator with the joint score.”
MethodApproximate Bayesian computation in population genetics
Beaumont, M. A., Zhang, W., & Balding, D. J. (2002) · Genetics · doi:10.1093/genetics/162.4.2025 · Cited by 2,932
Generative Adversarial Networks
(2022) · The Science of Deep Learning · doi:10.1017/9781108891530.013 · Cited by 2,402
IX. On the problem of the most efficient tests of statistical hypotheses
Neyman, J., & Pearson, E. S. (1933) · Philosophical Transactions of the Royal Society of London. Series A, Containing Papers of a Mathematical or Physical Character · doi:10.1098/rsta.1933.0009 · Cited by 2,341
Proceedings of the 19th International Conference on Artificial Intelligence and Statistics 2016
Murray, I., & Graham, M. M. (2016) · Cited by 2,115
Density estimation using Real NVP
Dinh, L., Sohl-Dickstein, J., & Bengio, S. (2016) · arXiv · Cited by 2,067
Conditional Image Generation with PixelCNN Decoders
van den Oord, A., Kalchbrenner, N., Vinyals, O., Espeholt, L., Graves, A., & Kavukcuoglu, K. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1606.05328 · Cited by 1,833
ON THE PROBLEM OF THE MOST EFFICIENT TESTS OF STATISTICAL HYPOTHESES
(1967) · Joint Statistical Papers · doi:10.1525/9780520339897-008 · Cited by 1,504
NICE: Non-linear Independent Components Estimation
Dinh, L., Krueger, D., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1410.8516 · Cited by 1,493 · cited 1×
“More recently, several methods for conditional density estimation have been proposed, often based on neural networks [ One particularly interesting class of models are normalizing flows [14–22], which model the density as a sequence of invertible transformations applied to a simple base density.”
MethodPixel Recurrent Neural Networks
van den Oord, A., Kalchbrenner, N., & Kavukcuoglu, K. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1601.06759 · Cited by 1,437 · cited 1×
“6), summing over x ∈ [5, 15], versus the training sample size.”
BackgroundBayesianly Justifiable and Relevant Frequency Calculations for the Applied Statistician
Rubin, D. B. (1984) · The Annals of Statistics · doi:10.1214/aos/1176346785 · Cited by 1,343
Variational Inference with Normalizing Flows
Rezende, D. J., & Mohamed, S. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1505.05770 · Cited by 1,253 · cited 1×
“6), summing over x ∈ [5, 15], versus the training sample size.”
BackgroundMarkov chain Monte Carlo without likelihoods
Marjoram, P., Molitor, J., Plagnol, V., & Tavaré, S. (2003) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.0306899100 · Cited by 1,199
The frontier of simulation-based inference
Cranmer, K., Brehmer, J., & Louppe, G. (2020) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1912789117 · Cited by 1,032
Self-citationAnalytical Note on Certain Rhythmic Relations in Organic Systems
Lotka, A. J. (1920) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.6.7.410 · Cited by 936 · cited 1×
“As a second example, we consider the Lotka-Volterra system [48, 49], a common example in the likelihood-free inference literature.”
MethodSimple Statistical Gradient-Following Algorithms for Connectionist Reinforcement Learning
Williams, R. J. (1992) · Reinforcement Learning · doi:10.1007/978-1-4615-3618-5_2 · Cited by 903
Sequential Monte Carlo without likelihoods
Sisson, S. A., Fan, Y., & Tanaka, M. M. (2007) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.0607208104 · Cited by 856
Simple statistical gradient-following algorithms for connectionist reinforcement learning
Williams, R. J. (1992) · Machine Learning · doi:10.1023/a:1022672621406 · Cited by 817
FFJORD: Free-form Continuous Dynamics for Scalable Reversible Generative Models
Grathwohl, W., Chen, R. T. Q., Bettencourt, J., Sutskever, I., & Duvenaud, D. (2018) · arXiv · doi:10.48550/arxiv.1810.01367 · Cited by 796 · cited 1×
“More recently, several methods for conditional density estimation have been proposed, often based on neural networks [ One particularly interesting class of models are normalizing flows [14–22], which model the density as a sequence of invertible transformations applied to a simple base density.”
MethodUNDAMPED OSCILLATIONS DERIVED FROM THE LAW OF MASS ACTION.
Lotka, A. J. (1920) · Journal of the American Chemical Society · doi:10.1021/ja01453a010 · Cited by 752 · cited 1×
“As a second example, we consider the Lotka-Volterra system [48, 49], a common example in the likelihood-free inference literature.”
MethodStan: A Probabilistic Programming Language for Bayesian Inference and Optimization
Gelman, A., Lee, D., & Guo, J. (2015) · Journal of Educational and Behavioral Statistics · doi:10.3102/1076998615606113 · Cited by 540
Pyro: Deep Universal Probabilistic Programming
Bingham, E., Chen, J. P., Jankowiak, M., Obermeyer, F. H., Pradhan, N., Karaletsos, T., Singh, R., Szerlip, P., Horsfall, P., & Goodman, N. D. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1810.09538 · Cited by 536 · cited 1×
“[43] we provide a framework that automates these calculations for any simulator in which all stochastic steps are implemented with the PYRO library [44].”
MethodHandbook of Approximate Bayesian Computation
(2018) · Chapman & Hall/CRC eBooks · doi:10.1201/9781315117195 · Cited by 450
MADE: Masked Autoencoder for Distribution Estimation
Germain, M., Gregor, K., Murray, I., & Larochelle, H. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1502.03509 · Cited by 448 · cited 1×
“For MAF and Scandal we stack four masked autoregressive distribution estimators (MADEs) [7] on a mixture of MADEs with 10 components.”
MethodMonte Carlo Methods of Inference for Implicit Statistical Models
Diggle, P. J., & Gratton, R. J. (1984) · Journal of the Royal Statistical Society Series B: Statistical Methodology · doi:10.1111/j.2517-6161.1984.tb01290.x · Cited by 427
Parameterized neural networks for high-energy physics
Baldi, P., Cranmer, K., Faucett, T., Sadowski, P., & Whiteson, D. (2016) · The European Physical Journal C · doi:10.1140/epjc/s10052-016-4099-4 · Cited by 231 · cited 1×
“We follow a second approach, in which we train parameterized estimators for the full model p̂(x|θ) or r̂(x|θ0, θ1) as a function of both the observables x and the parameters θ [9, 37].”
MethodSelf-citationA guide to constraining effective field theories with machine learning
Brehmer, J., Cranmer, K. S., Louppe, G., & Pavez, J. (2018) · Physical review. D/Physical review. D. · doi:10.1103/physrevd.98.052004 · Cited by 227
Self-citationLikelihood-Free Inference by Ratio Estimation
Thomas, O., Dutta, R., Corander, J., Kaski, S., & Gutmann, M. U. (2020) · Bayesian Analysis · doi:10.1214/20-ba1238 · Cited by 219
A new approach to probabilistic programming inference
Wood, F. (2014) · Open Collections · doi:10.14288/1.0044249 · Cited by 199
Masked Autoregressive Flow for Density Estimation
Papamakarios, G., Pavlakou, T., & Murray, I. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1705.07057 · Cited by 192 · cited 6×
“This “MAF MoG” architecture [22] performs better than a flow based on a single Gaussian base density.”
BackgroundMethodInfluentialMadMiner: Machine Learning-Based Inference for Particle Physics
Brehmer, J., Kling, F., Espejo, I., & Cranmer, K. S. (2020) · Computing and Software for Big Science · doi:10.1007/s41781-020-0035-2 · Cited by 191
Self-citationLikelihood-free inference with emulator networksPreprint
Lueckmann, J.-M., Bassetto, G., Karaletsos, T., & Macke, J. H. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1805.09294 · Cited by 188 · cited 1×
“They can be used to train a surrogate for the likelihood p(x|θ) [17, 19, 34] or, in a Bayesian setting, the posterior p(θ|x) [29, 33, 37].”
MethodLearning Disentangled Representations with Semi-Supervised Deep Generative ModelsPreprint
Siddharth, N., Paige, B., van de Meent, J.-W., Desmaison, A., Goodman, N. D., Kohli, P., Wood, F., & Torr, P. H. S. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1706.00400 · Cited by 181
Constraining Effective Field Theories with Machine Learning
Brehmer, J., Cranmer, K., Louppe, G., & Pavez, J. (2018) · Physical Review Letters · doi:10.1103/physrevlett.121.111801 · Cited by 160
Self-citationAlsing, J., Wandelt, B., & Feeney, S. (2018) · Monthly Notices of the Royal Astronomical Society · doi:10.1093/mnras/sty819 · Cited by 159 · cited 1×
“For inference in a sufficiently small neighborhood around a reference point θ ref , a precise estimator of the score ˆ t ( x | θ ref ) therefore defines a vector of ideal summary statistics that contain all the information in an observation x on the parameters θ [see also 3, 46, 47].”
BackgroundLearning in Implicit Generative Models
Mohamed, S., & Lakshminarayanan, B. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1610.03483 · Cited by 157 · cited 2×
“This “likelihood ratio trick” or “density ratio trick” is widely appreciated [9, 20, 26, 31].”
BackgroundFast $ε$-free Inference of Simulation Models with Bayesian Conditional Density Estimation
Papamakarios, G., & Murray, I. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1605.06376 · Cited by 133
Inference Compilation and Universal Probabilistic Programming
Le, T., Baydin, A. G., & Wood, F. (2016) · Oxford University Research Archive (ORA) (University of Oxford) · doi:10.48550/arxiv.1610.09900 · Cited by 131 · cited 1×
“Probabilistic programming systems represent another class of methods that use the simulator directly during inference [29, 30].”
BackgroundFlexible statistical inference for mechanistic models of neural dynamics
Lueckmann, J.-M., Gonçalves, P. J., Bassetto, G., Öcal, K., Nonnenmacher, M., & Macke, J. H. (2017) · Lirias · doi:10.48550/arxiv.1711.01861 · Cited by 129 · cited 1×
“They can be used to train a surrogate for the likelihood p(x|θ) [17, 19, 34] or, in a Bayesian setting, the posterior p(θ|x) [29, 33, 37].”
MethodApproximating Likelihood Ratios with Calibrated Discriminative Classifiers
Cranmer, K. S., Pavez, J., & Louppe, G. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1506.02169 · Cited by 127 · cited 4×
“This “likelihood ratio trick” or “density ratio trick” is widely appreciated [9, 20, 26, 31].”
MethodInfluentialSelf-citationLikelihood-free MCMC with Amortized Approximate Ratio EstimatorsPreprint
Hermans, J., Begy, V., & Louppe, G. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1903.04057 · Cited by 125
Self-citationNuisance hardened data compression for fast likelihood-free inference
Alsing, J., & Wandelt, B. (2019) · Monthly Notices of the Royal Astronomical Society · doi:10.1093/mnras/stz1900 · Cited by 116
Automatic physical inference with information maximizing neural networks
Charnock, T., Lavaux, G., & Wandelt, B. D. (2018) · Physical Review D · doi:10.1103/physrevd.97.083004 · Cited by 113 · cited 1×
“…population genetics, which has motivated the development of likelihood-free inference algorithms such as Approximate Bayesian Computation (ABC) [1–4] and neural density estimation (NDE) techniques [ We present a suite of new techniques that can dramatically improve the sample efficiency for…”
MethodDeep Probabilistic Programming
Tran, D., Hoffman, M. D., Saurous, R. A., Brevdo, E., Murphy, K., & Blei, D. M. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1701.03757 · Cited by 107
Generalized massive optimal data compression
Alsing, J., & Wandelt, B. (2018) · Monthly Notices of the Royal Astronomical Society: Letters · doi:10.1093/mnrasl/sly029 · Cited by 100 · cited 1×
“For inference in a sufficiently small neighborhood around a reference point θ ref , a precise estimator of the score ˆ t ( x | θ ref ) therefore defines a vector of ideal summary statistics that contain all the information in an observation x on the parameters θ [see also 3, 46, 47].”
BackgroundLikelihood-free inference via classification
Gutmann, M. U., Dutta, R., Kaski, S., & Corander, J. (2018) · Statistics and Computing · doi:10.1007/s11222-017-9738-6 · Cited by 90 · cited 5×
“Most established inference methods, such as ABC and techniques based on neural density estimation, only use samples of observables from the simulator.”
MethodInfluentialHierarchical Implicit Models and Likelihood-Free Variational Inference
Tran, D., Ranganath, R., & Blei, D. M. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1702.08896 · Cited by 86
Brehmer, J., Mishra-Sharma, S., Hermans, J., Louppe, G., & Cranmer, K. (2019) · The Astrophysical Journal · doi:10.3847/1538-4357/ab4c41 · Cited by 80
Self-citationApproximate Bayesian computation via regression density estimation
Fan, Y., Nott, D. J., & Sisson, S. A. (2013) · Stat · doi:10.1002/sta4.15 · Cited by 78 · cited 1×
“6), summing over x ∈ [5, 15], versus the training sample size (which refers to the total number of x samples, distributed over 10 values of θ ∈ [−1,−0.”
BackgroundSequential Neural Likelihood: Fast Likelihood-free Inference with Autoregressive Flows
Papamakarios, G., Sterratt, D. C., & Murray, I. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1805.07226 · Cited by 67
Adversarial Variational Optimization of Non-Differentiable SimulatorsPreprint
Louppe, G., Hermans, J., & Cranmer, K. S. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1707.07113 · Cited by 57 · cited 2×
“This poses a difficulty, making methods based on ∇ z x [42] and ∇ θ x inapplicable, which previously motivated techniques such as Adversarial Variational Optimization [32].”
MethodSelf-citationInference Networks for Sequential Monte Carlo in Graphical Models
Paige, B., & Wood, F. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1602.06701 · Cited by 46
Dynamic Likelihood-free Inference via Ratio Estimation (DIRE)Preprint
Dinev, T., & Gutmann, M. U. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1810.09899 · Cited by 26
Asymptotically exact inference in differentiable generative models
Graham, M. M., & Storkey, A. (2017) · Electronic Journal of Statistics · doi:10.1214/17-ejs1340si · Cited by 24
Neural Autoregressive Distribution Estimation
Uría, B., Côté, M.-A., Gregor, K., Murray, I., & Larochelle, H. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1605.02226 · Cited by 19 · cited 1×
“6), summing over x ∈ [5, 15], versus the training sample size (which refers to the total number of x samples, distributed over 10 values of θ ∈ [−1,−0.”
Backgroundcarl: a likelihood-free inference toolbox
Louppe, G., Cranmer, K. S., & Pavez, J. (2016) · The Journal of Open Source Software · doi:10.21105/joss.00011 · Cited by 16 · cited 2×
“lihood ratio): The same trick can be used to improve the likelihood ratio trick and the Carl inference method [9, 32].”
MethodSelf-citationUnifying Generative Models And Exact Likelihood-Free Inference With Conditional Bijections
Cranmer, K. S., & Louppe, G. (2016) · Zenodo (CERN European Organization for Nuclear Research) · doi:10.5281/zenodo.198541 · Cited by 12
Self-citationConstraining effective field theories with machine learning
Brehmer, J., Cranmer, K. S., Espejo, I., Held, A., Kling, F., Louppe, G., & Pavez, J. (2020) · EPJ Web of Conferences · doi:10.1051/epjconf/202024506026 · Cited by 9
Self-citation
Cited by
Machine learning and the physical sciences
Carleo, G., Cirac, I., Cranmer, K., Daudet, L., Schuld, M., Tishby, N., Vogt-Maranto, L., & Zdeborová, L. (2019) · Reviews of Modern Physics · doi:10.1103/revmodphys.91.045002 · Cited by 2,417
Self-citationThe frontier of simulation-based inference
Cranmer, K., Brehmer, J., & Louppe, G. (2020) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1912789117 · Cited by 1,032
Self-citationLearning new physics from a machine
D’Agnolo, R. T., & Wulzer, A. (2019) · Physical Review D · doi:10.1103/physrevd.99.015014 · Cited by 203
OmniFold: A Method to Simultaneously Unfold All Observables
Andreassen, A., Komiske, P. T., Metodiev, E. M., Nachman, B., & Thaler, J. (2020) · Physical Review Letters · doi:10.1103/physrevlett.124.182001 · Cited by 164
“[22] Johann Brehmer, Gilles Louppe, Juan Pavez, and Kyle Cranmer, “Mining gold from implicit models to improve likelihood-free inference,” (2018), arXiv:1805.”
BackgroundBrehmer, J., Mishra-Sharma, S., Hermans, J., Louppe, G., & Cranmer, K. (2019) · The Astrophysical Journal · doi:10.3847/1538-4357/ab4c41 · Cited by 80
Self-citation
Metadata
- Authors
- Brehmer, J., Louppe, G., Pavez, J., & Cranmer, K.
- Venue
- Proceedings of the National Academy of Sciences
- Published
- 2020-02-20
- Volume
- 117
- Issue
- 10
- Pages
- 5242-5249
- ISSN
- 0027-8424 (ISSN-L)
- 1091-6490
- OpenAlex
- W3006943238
- Cited by
- 159
- References
- 64
- FWCI
- 15.89
- Type
- journal article
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.