IP Library › Granted Patent US 12,437,194
Granted Patent B2
US 12,437,194 · App. 17/360,666 · Granted Oct 7, 2025

Mechanistic model parameter inference through artificial intelligence

Inventors: Viatcheslav Gurev (Bedford Hills, NY); James R. Kozloski (New Fairfield, CT); Kenney Ng (Arlington, MA); Jaimit Parikh (Danbury, CT)
Assignee: INTERNATIONAL BUSINESS MACHINES CORPORATION
G06N3/08G06N3/045G06N3/094
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,437,194
App. No.
17/360,666
Granted
Oct 7, 2025
Kind
B2
Abstract

Techniques regarding inferring parameters of one or more mechanistic models are provided. For example, one or more embodiments described herein can comprise a system, which can comprise a memory that can store computer executable components. The system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can comprise a machine learning component that can identify a causal relationship in a mechanistic model via a machine learning architecture that employs a parameter space of the mechanistic model as a learned distribution sampled within a generative adversarial network.

Claims (37)

1. A system, comprising:

a memory that stores computer executable components; and

a processor that executes at least one of the computer executable components that:

constructs a machine learning architecture comprising:

an encoder layer of the machine learning architecture, wherein the encoder layer comprises at least one encoder and at least one machine learning network, and wherein the at least one machine learning network comprises at least one generative adversarial network;

an intermediate layer of the machine learning architecture that is communicatively coupled to the encoder layer, wherein the intermediate layer comprise at least one bijector node; and

a decoder layer of the machine learning architecture that is communicatively coupled to the intermediate layer, wherein the decoder layer comprises a mechanistic model that operates as a decoder of the machine learning architecture, and wherein the at least one bijector node transforms a Gaussian distribution generated by the at least one machine learning network to a prior distribution of model parameters of the mechanistic model inputted to the at least one machine learning network; and

trains the machine learning architecture to identify a causal relationship in the mechanistic model using a parameter space of the mechanistic model as a learned distribution sampled within the at least one generative adversarial network based on a sample subset of outputs of the mechanistic model.

2. The system of claim 1 , wherein the at least one generative adversarial network is selected from a group consisting of a conditional generative adversarial network, a regularized generative adversarial network, and a conditional regularized generative adversarial network.

3. The system of claim 1 , wherein the at least one generative adversarial network determines a conditional probability associated with the parameter space based on an output of the mechanistic model.

4. The system of claim 1 , wherein the machine learning architecture approximates a distribution of the parameter space that is consistent with a single output of the mechanistic model or coherent with a distribution of outputs of the mechanistic model.

5. The system of claim 1 , wherein the parameter space is associated with the output and is known.

6. The system of claim 1 , wherein the at least one generative adversarial network is a regularized generative adversarial network or a conditional regularized generative adversarial network that comprises at least two discriminator nodes, and wherein a generator node loss function of the regularized generative adversarial network or the conditional regularized generative adversarial network is a weighted sum of losses due to the at least two discriminator nodes.

7. The system of claim 1 , wherein the at least one generative adversarial network is a regularized generative adversarial network or a conditional regularized generative adversarial network that generates an input parameter for the mechanistic model which, when forwarded through the mechanistic model, results in mechanistic model outputs that are coherent with a set of observation data characterized by the mechanistic model.

8. The system of claim 7 , wherein the input parameter is conditioned on auxiliary variables derived from a source other than the prior distribution accessible to the mechanistic model.

9. The system of claim 1 , wherein the mechanistic model is at least of a biophysical model of a biological system or a trained surrogate of the biophysical model of the biological system.

10. The system of claim 9 , wherein the parameter space characterizes observations of the biological system.

11. A computer-implemented method, comprising:

constructing, by at least one processor of a system, a machine learning architecture comprising:

an encoder layer of the machine learning architecture, wherein the encoder layer comprises at least one encoder and at least one machine learning network, and wherein the at least one machine learning network comprises at least one generative adversarial network;

an intermediate layer of the machine learning architecture that is communicatively coupled to the encoder layer, wherein the intermediate layer comprise at least one bijector node; and

a decoder layer of the machine learning architecture that is communicatively coupled to the intermediate layer, wherein the decoder layer comprises a mechanistic model that operates as a decoder of the machine learning architecture, and wherein the at least one bijector node transforms a Gaussian distribution generated by the at least one machine learning network to a prior distribution of model parameters of the mechanistic model inputted to the at least one machine learning network; and

training, by the at least one processor, the machine learning architecture to identify a causal relationship in the mechanistic model using a parameter space of the mechanistic model as a learned distribution sampled within the at least one generative adversarial network based on a sample subset of outputs of the mechanistic model.

12. The computer-implemented method of claim 11 , wherein the at least one generative adversarial network is selected from a group consisting of a conditional generative adversarial network, a regularized generative adversarial network, and a conditional regularized generative adversarial network.

13. The computer-implemented method of claim 11 , wherein the machine learning architecture a distribution of the parameter space that is consistent with a single output of the mechanistic model or coherent with a distribution of outputs of the mechanistic model.

14. The computer-implemented method of claim 11 , wherein the parameter space is associated with the output and is known.

15. The computer-implemented method of claim 11 , wherein the at least one generative adversarial network is a regularized generative adversarial network or conditional regularized generative adversarial network that generates an input parameter for the mechanistic model that, when forwarded through the mechanistic model, results in mechanistic model outputs that are coherent with a set of observation data characterized by the mechanistic model.

16. A computer program product for autonomous model parameter inferencing, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:

construct a machine learning architecture comprising:

an encoder layer of the machine learning architecture, wherein the encoder layer comprises at least one encoder and at least one machine learning network, and wherein the at least one machine learning network comprises at least one generative adversarial network;

an intermediate layer of the machine learning architecture that is communicatively coupled to the encoder layer, wherein the intermediate layer comprise at least one bijector node; and

a decoder in a decoder layer of the machine learning architecture that is communicatively coupled to the intermediate layer, wherein the decoder layer comprises a mechanistic model that operates as a decoder of the machine learning architecture, and wherein the at least one bijector node transforms a Gaussian distribution generated by the at least one machine learning network to a prior distribution of model parameters of the mechanistic model inputted to the at least one machine learning network; and

train the machine learning architecture to identify a causal relationship in the mechanistic model using a parameter space of the mechanistic model as a learned distribution sampled within the at least one generative adversarial network based on a sample subset of outputs of the mechanistic model.

17. The computer program product of claim 16 , wherein the at least one generative adversarial network is selected from a group consisting of a conditional generative adversarial network and a regularized generative adversarial network or conditional regularized generative adversarial network.

18. The computer program product of claim 16 , wherein the at least one generative adversarial network determines a conditional probability associated with the parameter space based on an output of the mechanistic model.

19. The computer program product of claim 16 , wherein the at least one generative adversarial network is a regularized generative adversarial network or a conditional regularized generative adversarial network that generates an input parameter for the mechanistic model that, when forwarded through the mechanistic model, results in mechanistic model outputs that are coherent with a set of observation data characterized by the mechanistic model.

20. The computer program product of claim 16 , wherein the mechanistic model is a biophysical model of a biological system.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jun 28, 2021
From: GUREV, VIATCHESLAV; KOZLOSKI, JAMES R.; NG, KENNEY; PARIKH, JAIMIT
To: INTERNATIONAL BUSINESS MACHINES CORPORATION
Reel/Frame 056690/0983 →
Continuity (1)
Related Publication 20220414452A1 · Dec 29, 2022
References Cited (39)
US 11174289B1 · Das et al. · 2021 [cited by applicant]
US 20150294218A1 · Thomson et al. · 2015 [cited by applicant]
US 20190026869A1 · Gupta · 2019 [cited by examiner]
US 20200233920A1 · Meeds · 2020 [cited by examiner]
US 20200342359A1 · Hu et al. · 2020 [cited by applicant]
US 20220414451A1 · Gurev et al. · 2022 [cited by applicant]
US 20230045548A1 · Yakut et al. · 2023 [cited by applicant]
Xu, K., & Darve, E. (2021). Solving inverse problems in stochastic models using deep neural networks and adversarial training. Computer Methods in Applied Mechanics and Engineering, 384, 113976. (Year: 2021). [cited by examiner]
Mirza, Mehdi, and Simon Osindero. “Conditional generative adversarial nets.” arXiv preprint arXiv:1411.1784 (2014). (Year: 2014). [cited by examiner]
Nguyen, Tu, et al. “Dual discriminator generative adversarial nets.” Advances in neural information processing systems 30 (2017). ( Year: 2017). [cited by examiner]
Berthelot, David. “BEGAN: Boundary Equilibrium Generative Adversarial Networks.” arXiv preprint arXiv:1703.10717 (2017) (Year: 2017). [cited by examiner]
McMahon, et al., “A general mechanistic model enables predictions of the biological effectiveness of different qualities of radiation,” Published Online Sep. 7, 2017, Scientific Reports | 7: 10790 | DOI:10.1038/s41598-0… [cited by applicant]
Stites, “Chemical kinetic mechanistic models to investigate cancer biology and impact cancer medicine,” Publication Info: Physical Biology 10.2: 026004 (6 .). IOP Publishing Ltd. (Apr. 2013), 7 pages. [cited by applicant]
Hahn, “Solving Inverse Problems for Mechanistic Systems Biology Models with Unknown Inputs. Final Report for Jan. 1-Sep. 30, 2014,” Publication info: National Technical Information Service . (Oct. 16, 2014), 31 pages. [cited by applicant]
Srivastava, et al., “VEEGAN: Reducing Mode Collapse in GANs using Implicit Variational Learning,” arXiv:1705.07761v3 [stat.ML] Nov. 6, 2017, 17 pages. [cited by applicant]
Parikh, et al., “Integration of AI and Mechanistic Modeling in Generative Adversarial Networks for Stochastic Inverse Problems,” Jan. 29, 2021, 19 pages. [cited by applicant]
Kingma, et al., Improving Variational Inference with Inverse Autoregressive Flow. arXiv:1606.04934v2 [cs.LG] Jan. 30, 2017, https://arxiv.org/abs/1606.04934, 16 pages. [cited by applicant]
Oord, et al., “Parallel WaveNet: Fast High-Fidelity Speech Synthesis,” arXiv:1711.10433v1 [cs.LG] Nov. 28, 2017, https://arxiv.org/abs/1711.10433, 11 pages. [cited by applicant]
Poole, et al., “Inference for Deterministic Simulation Models The Bayesian Melding Approach,” Technical Report No. 346, Department of Statistics, University of Washington, Dec. 1998, 57 pages. [cited by applicant]
Papamakarios, et al., Masked Autoregressive Flow for Density Estimation. rXiv:1705.07057v4 [stat.ML] Jun. 14, 2018, https://arxiv.org /abs/1705.07057, 17 pages. [cited by applicant]
Poole, et al., Inference for Deterministic Simulation Models: The Bayesian Melding Approach. Poole and Raftery 2000. https://www.tandfonline.com/doi/abs/10.1080/01621459.2000.10474324, Journal of the American Statistica… [cited by applicant]
Mell, et al., “The NIST Definition of Cloud Computing,” Special Publication 800-145, Recommendations of the National Institute of Standards and Technology, Sep. 2011, 7 pages. [cited by applicant]
List of IBM Patents or Applications Treated as Related. [cited by applicant]
Neyshabur, et al., “Stabilizing GAN Training with Multiple Random Projections,” arXiv:1705.07831v2 [cs.LG] Jun. 23, 2018, 21 pages. [cited by applicant]
Belghazi, et al., “Mutual Information Neural Estimation,” arXiv: 1801.04062v4 [cs.LG] Jun. 7, 2018, 5 pages. [cited by applicant]
Albuquerque, et al., “Multi-objective training of Generative Adversarial Networks with multiple discriminators,” arXiv:1901.08680v1 [cs.LG] Jan. 24, 2019, 16 pages. [cited by applicant]
Wiatrak, et al., “Stabilizing Generative Adversarial Networks: A Survey,” arXiv:1910.00927v2 [cs.LG] Mar. 24, 2020, 26 pages. [cited by applicant]
Huang, et al., “Augmented Normalizing Flows: Bridging the Gap Between Generative Flows and Latent Variable Models,” arXiv:2002.07101v1 [cs.LG] Feb. 17, 2020, 27 pages. [cited by applicant]
Parikh, et al., “Integration of AI and Mechanistic Modeling in Generative Adversarial Networks for Stochastic Inverse Problems,” arXiv:2009.08267v2 [stat.ML] Oct. 27, 2020, 19 pages. [cited by applicant]
Fengler, et al., “Likelihood Approximation Networks (LANs) for Fast Inference of Simulation Models in Cognitive Neuroscience, ” bioRxiv preprint doi: https://doi.org/10.1101/2020.11.20.392274; this version posted Nov. 2… [cited by applicant]
Sugiyama, et al., “Density-ratio matching under the Bregman divergence: a unified framework of density-ratio estimation,” Ann Inst Stat Math (2012) 64:1009-1044 DOI 10.1007/s10463-011-0343-8, 6 pages. [cited by applicant]
Jonsson, et al., “Nonlinearity Detection: Advantages of Nonlinear Mixed-Effects Modeling,” AAPS PharmSci 2000; 2(3) article 32 (http://www.pharmsci.org), 10 pages. [cited by applicant]
Berard, et al., “Embedding Riemannian Manifolds by their Heat Kernel,” Geometric and Functional Analysis, vol. 4, No. 4 (1994), 26 pages. [cited by applicant]
Belkin, et al., “Using Manifold Structure for Partially Labelled Classification,” NIPS'02: Proceedings of the 15th International Conference on Neural Information Processing Systems, 8 pages. [cited by applicant]
Coifman, et al., “Geometric diffusions as a tool for harmonic analysis and structure definition of data: Diffusion maps,” pp. 7426-7431—PNAS—May 24, 2005—vol. 102—No. 21. [cited by applicant]
Zitzler, et al., “Indicator-Based Selection in Multiobjective Search,” Proceedings of the 8th International Conference on Parallel Problem Solving from Nature (PPSN VIII) Sep. 2004, Birmingham, UK, 10 pages. [cited by applicant]
Wang et al., “Deep generative modeling for mechanistic-based learning and design of metamaterial systems”, Computer Methods in Applied Mechanics and Engineering, Sep. 7, 2020, 23 pages. [cited by applicant]
Goodfellow et al., “Generative Adversarial Nets”, arXiv: 1406.2661v1 [stat. ML], Jun. 10, 2014, 9 pages. [cited by applicant]
Mecheder et al., “Adversarial variational bayes: Unifying variational autoencoders and generative adversarial networks”, arXiv: 1701.04722v4 [cs.LG], Aug. 6, 2017, 10 pages. [cited by applicant]