Optimizing machine learning decisions with prediction uncertainty

Digital background image of brain connectors
Digital background depicting innovative technologies in (AI) artificial systems, neural interfaces and internet machine learning technologies

While ML classifiers are widespread, output is often not part of a follow-on decision-making process because of lack of uncertainty quantification. Through this project, the team developed decision analysis methods that combined uncertainty estimates for ML predictions with a domain-specific model of error costs. In the project, they explicitly weighed whether ML models under evaluation were qualified to make any prediction by producing general algorithms that minimized prediction error costs by validating these algorithms through their demonstrations on cyber security and image analysis cases.

The developed and trained ML classifier ultimately provided a framework for: (1) quantifying and propagating uncertainty in ML classifiers; (2) formally linking ML outputs with the decision-making process; and (3) making optimal decisions for classification under uncertainty with single or multiple objectives. Methods developed through this project are currently being incorporated into applications that impact national security domains by directly addressing questions of automated decision.

Sandia researchers linked to work

  • Michael Darling
  • Justin Doak

Sponsored by

Image of LDRD_Icon-01

Associated Pubications

  • Doak, J., Darling, M., & Darling, M. (2022). Preliminary Results for Using Uncertainty and Out-of-distribution Detection to Identify Unreliable Predictions. https://doi.org/10.2172/1899654 Publication ID: 80438
  • Field, R., Darling, M., & Darling, M. (2022). A Decision Theoretic Approach To Optimizing Machine Learning Decisions with Prediction Uncertainty. https://doi.org/10.2172/1899419 Publication ID: 80436
  • Leger, M., Darling, M., Jones, S., Matzen, L., Stracuzzi, D., Wilson, A., Bueno, D., Christentsen, M., Ginaldi, M., Hannasch, D., Heidbrink, S., Howell, B., Leger, C., Reedy, G., Rogers, A., Williams, J., & Williams, J. (2021). Exploring Explicit Uncertainty for Binary Analysis (EUBA). https://doi.org/10.2172/1832314 Publication ID: 76829
  • Darling, M., Doak, J., Field, R., Smith, M., & Smith, M. (2021). Optimizing Machine Learning Decisions with Prediction Uncertainty [Conference Presenation]. https://doi.org/10.2172/1888406 Publication ID: 79403
  • Smith, Z., Darling, M., Stracuzzi, D., Doak, J., Smith, M., Bickel, J., & Bickel, J. (2021). Optimizing Machine Learning Predictions with Prediction Uncertainty [Conference Poster]. https://doi.org/10.2172/1860307 Publication ID: 77800
  • Doak, J. (2020). I could see your lips move [Presentation]. https://www.osti.gov/biblio/1804992 Publication ID: 73931
  • Doak, J., Smith, M., Ingram, J., & Ingram, J. (2020). Self-Updating Models with Error Remediation [Conference Poster]. https://doi.org/10.1117/12.2563843 Publication ID: 73520
  • Hayden, N., Reinhardt, J., Stewart, M., Doak, J., Gunda, T., Abel, K., & Abel, K. (2020). Artificial Intelligence and Autonomy in Space: Balancing Risks of Unintended Escalation [Conference Poster]. https://www.osti.gov/biblio/1780585 Publication ID: 73350
  • Darling, M., Hush, D., Stracuzzi, D., & Stracuzzi, D. (2020). A Novel Measure of Uncertainty For Machine Learning Predictions [Conference Poster]. https://www.osti.gov/biblio/1768156 Publication ID: 72788
  • Darling, M. (2019). Using Uncertainty To Interpret Supervised Machine Learning Predictions. https://www.osti.gov/biblio/1592876 Publication ID: 66905
  • Doak, J., Ingram, J., Smith, M., Vineyard, C., & Vineyard, C. (2019). Self-updating Models with Error Remediation for Bandwidth-constrained Environments [Conference Poster]. https://www.osti.gov/biblio/1641249 Publication ID: 69923
  • Doak, J. (2019). I could see your lips move [Presentation]. https://www.osti.gov/biblio/1644871 Publication ID: 67996
  • Doak, J., Ingram, J., Mulder, S., Naegle, J., Cox, J.A., Aimone, J.B., DIxon, K.R., James, C., Follett, D.R., & Follett, D.R. (2018). Tracking Cyber Adversaries with Adaptive Indicators of Compromise [Conference Poster]. Proceedings – 2017 International Conference on Computational Science and Computational Intelligence, CSCI 2017. https://doi.org/10.1109/CSCI.2017.2 Publication ID: 54611
  • Stracuzzi, D., Darling, M., Peterson, M., Chen, M., & Chen, M. (2018). Quantifying Uncertainty to Improve Decision Making in Machine Learning. https://doi.org/10.2172/1481629 Publication ID: 59354
  • Chen, M., Darling, M., Stracuzzi, D., & Stracuzzi, D. (2018). Preliminary Results on Applying Nonparametric Clustering and Bayesian Consensus Clustering Methods to Multimodal Data. https://doi.org/10.2172/1475256 Publication ID: 59212
  • Smith, M., Ingram, J., Lamb, C., Draelos, T., Doak, J., Aimone, J.B., James, C., & James, C. (2018). Dynamic Analysis of Executables to Detect and Characterize Malware [Conference Poster]. https://doi.org/10.1109/ICMLA.2018.00011 Publication ID: 59291
  • Chen, M., Stracuzzi, D., Darling, M., & Darling, M. (2018). A Mathematical Framework for Uncertainty Quantification in Multimodal Image Analysis via Probabilistic Clustering Models [Conference Poster]. https://www.osti.gov/biblio/1591721 Publication ID: 63998
  • Chen, M., Stracuzzi, D., Darling, M., & Darling, M. (2018). A Mathematical Framework for Uncertainty Quantification in Multimodal Image Analysis via Probabilistic Clustering Models [Conference Poster]. https://www.osti.gov/biblio/1575046 Publication ID: 63512
  • Doak, J., Coon, J., Fernandez, R., Louie, M., Stangebye, T., & Stangebye, T. (2018). Capsule Networks: Capturing Presence and Orientation of Representations [Conference Poster]. https://www.osti.gov/biblio/1573310 Publication ID: 63439
  • Chen, M., Darling, M., Vollmer, C., Peterson, M., Stracuzzi, D., & Stracuzzi, D. (2018). Using Uncertainty to Understand Machine Learning Models and Decisions [Conference Poster]. https://www.osti.gov/biblio/1526113 Publication ID: 62451
  • Stracuzzi, D., Darling, M., Chen, M., Peterson, M., & Peterson, M. (2018). Data-Driven Uncertainty Quantification for Multi-Sensor Analytics [Conference Poster]. https://www.osti.gov/biblio/1507484 Publication ID: 61468
  • Darling, M., Stracuzzi, D., Chen, M., & Chen, M. (2018). Uncertainty Propagation In Multilayer Analysis [Conference Poster]. https://www.osti.gov/biblio/1498629 Publication ID: 60943
  • Darling, M., Luger, G.F., Jones, T.B., Denman, M., Groth, K.M., & Groth, K.M. (2018). Intelligent Modeling for Nuclear Power Plant Accident Management. International Journal on Artificial Intelligence Tools, 27(2). https://doi.org/10.1142/S0218213018500033 Publication ID: 58554
  • Chen, M., Darling, M., Stracuzzi, D., & Stracuzzi, D. (2018). Multimodal Image Analysis and Uncertainty Quantification via Nonparametric Probabilistic Clustering [Conference Poster]. https://www.osti.gov/biblio/1498444 Publication ID: 60902
  • Stracuzzi, D., Darling, M., Chen, M., Peterson, M., & Peterson, M. (2018). Data-driven uncertainty quantification for multisensor analytics [Conference Poster]. Proceedings of SPIE – The International Society for Optical Engineering. https://doi.org/10.1117/12.2304921 Publication ID: 61586
  • Ingram, J., Draelos, T., Sahakian, M., Doak, J., & Doak, J. (2017). Temporal Cyber Attack Detection. https://doi.org/10.2172/1409921 Publication ID: 54278
  • Chen, M., Stracuzzi, D., Darling, M., Peterson, M., & Peterson, M. (2017). Establishing Uniform Image Segmentation Ground Truth Protocols for Uncertainty Quantification and Improved Model Evaluation [Conference Poster]. https://www.osti.gov/biblio/1511807 Publication ID: 54356
  • Doak, J., Doak, J., Ingram, J., Ingram, J., Mulder, S., Mulder, S., Naegle, J., Naegle, J., Cox, J., Cox, J., Aimone, J.B., Aimone, J.B., DIxon, K.R., DIxon, K.R., James, C., James, C., Follet, D., Follet, D., & Follet, D. (2017). Tracking Cyber Adversaries with Adaptive Indicators of Compromise [Conference Poster]. https://doi.org/10.1109/CSCI.2017.2 Publication ID: 54191
  • Stracuzzi, D., Chen, M., Darling, M., Peterson, M., & Peterson, M. (2017). From Data to Decisions: Placing Machine Learning Challenges In Context [Conference Poster]. https://www.osti.gov/biblio/1484938 Publication ID: 54484
  • Stracuzzi, D., Chen, M., Darling, M., Dauphin, S., Peterson, M., Vollmer, C., & Vollmer, C. (2017). Data-Driven Uncertainty Quantification for Remote Sensing [Presentation]. https://www.osti.gov/biblio/1457960 Publication ID: 56281
  • Stracuzzi, D., Chen, M., Darling, M., Dauphin, S., Peterson, M., Vollmer, C., Young, C.J., & Young, C.J. (2016). Uncertainty in Data Analytics [Presentation]. https://www.osti.gov/biblio/1413588 Publication ID: 48176
  • Darling, M., Groth, K.M., & Groth, K.M. (2016). A Dynamic Bayesian Network for Diagnosing Nuclear Power Plant Accidents [Presentation]. https://www.osti.gov/biblio/1375590 Publication ID: 51647
  • Doak, J. (2016). Cyber Training Curriculum [Presentation]. https://www.osti.gov/biblio/1514625 Publication ID: 51375
  • Darling, M., Groth, K.M., Denman, M., Jones, T., Luger, G., & Luger, G. (2016). A Dynamic Bayesian Network for Diagnosing Nuclear Power Plant Accidents [Conference Poster]. https://www.osti.gov/biblio/1367135 Publication ID: 49938
  • Bierma, M., Doak, J., Hudson, C.M., & Hudson, C.M. (2016). Learning to Rank for Alert Triage [Conference Poster]. https://doi.org/10.1109/THS.2016.7568907 Publication ID: 49871
  • Bierma, M., Doak, J., Hudson, C.M., & Hudson, C.M. (2016). Learning to Rank for Alert Triage [Conference Poster]. https://doi.org/10.1109/THS.2016.7568907 Publication ID: 49371
  • Jones, T.B., Darling, M., Groth, K.M., Denman, M.R., Luger, G.F., & Luger, G.F. (2016). A dynamic Bayesian network for diagnosing nuclear power plant accidents [Conference Poster]. Proceedings of the 29th International Florida Artificial Intelligence Research Society Conference, FLAIRS 2016. https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85003944802&origin=inward Publication ID: 48503
  • Johnson, J., Doak, J., Ingram, J., Shelburg, J., & Shelburg, J. (2013). Active Learning for Alert Triage [Conference]. https://www.osti.gov/biblio/1121155 Publication ID: 31722
  • Doak, J., Ingram, J., Johnson, J., Shelburg, J., & Shelburg, J. (2013). Active Learning for Alert Triage [Conference]. https://www.osti.gov/biblio/1115032 Publication ID: 35198
  • Suppona, R.A., Wilson, A., Doak, J., & Doak, J. (2013). Can we identify spear phishing targets before the email is sent? [Conference]. https://www.osti.gov/biblio/1115637 Publication ID: 33496
  • Haas, J., Doak, J., Crosby, S., Helinski, R., Lamb, C., & Lamb, C. (2013). Dynamic defense workshop :. https://doi.org/10.2172/1093703 Publication ID: 32106
  • Suppona, R.A., Doak, J., & Doak, J. (2012). LDRD Annual Report blurb for Jeremy Wendt’s LDRD [Presentation]. https://www.osti.gov/biblio/1658291 Publication ID: 30318
  • Haas, J., Doak, J., Hamlet, J., & Hamlet, J. (2012). Machine-Oriented Biometrics and Cocooning for Dynamic Network Defense [Conference]. https://www.osti.gov/biblio/1116590 Publication ID: 29628
  • Doak, J., Wilson, A., Suppona, R.A., & Suppona, R.A. (2012). Identifying Dynamic Patterns in Network Traffic to Predict and Mitigate Cyberattacks [Presentation]. https://www.osti.gov/biblio/1658713 Publication ID: 29028
Showing 10 of 44 publications.

May 9, 2023