Deep causal learning for continuous testing, diagnosis, and optimization
A system and methods for multivariant learning and optimization repeatedly generate self-organized experimental units (SOEUs) based on the one or more assumptions for a randomized multivariate comparison of process decisions to be provided to users of a system. The SOEUs are injected into the system to generate quantified inferences about the process decisions. Responsive to injecting the SOEUs, at least one confidence interval is identified within the quantified inferences, and the SOEUs are iteratively modified based on the at least one confidence interval to identify at least one causal interaction of the process decisions within the system. The causal interaction can be used for testing, diagnosis, and optimization of the system performance.
1. A system for multivariate learning and optimization, comprising:
memory; and
a processor coupled to the memory, the processor configured to:
receive one or more assumptions for a randomized multivariate comparison of process decisions, the process decisions to be provided to users of a system;
repeatedly generate self-organized experimental units (SOEUs) based on the one or more assumptions;
inject the SOEUs into the system to generate quantified inferences about the process decisions;
identify, responsive to injecting the SOEUs, at least one confidence interval within the quantified inferences;
compute the at least one confidence interval by statistical testing on d-scores, which are defined as differences between measured effects when a variable is activated and when it is deactivated; and
iteratively modify the SOEUs based on the at least one confidence interval to identify at least one causal interaction of the process decisions within the system.
2. The system of claim 1 , wherein the causal interaction is used for testing and improving upon combinatoric states of ordinal and temporal process controls.
3. The system of claim 1 , wherein the causal interaction is used for queueing system demands and allocating responses.
4. The system of claim 1 , wherein the causal interaction is used for testing and improving sensor network topography, calibration and interpretation.
5. The system of claim 1 , wherein the SOEU duration and other characteristics are perturbed over time to minimize temporal and spatial interactions and maximize independence of SOEUs.
6. The system of claim 1 , wherein SOEU characteristics are used to optimize cluster assignment and improve exchangeability of SOEUs within each cluster.
7. The system of claim 1 , wherein an amount of data used to generate quantified inferences (DIW) is proportional to the stability of the causal interaction in time.
8. The system of claim 7 , wherein the DIW is less than a total amount of data available by 10%, 20%, 40%, or 80%.
9. The system of claim 1 , wherein an objective function of the system changes over space and/or time.
10. The system of claim 1 , wherein random assignments are used for baseline monitoring of the system.
11. The system of claim 10 , wherein the baseline monitoring is used to objectively tune or optimize hyperparameters of the system.
12. The system of claim 1 , wherein the confidence interval provides an unbiased estimate of physical cause and effect relationships.
13. The system of claim 1 , wherein
clustering and blocking are part of how the experimental units are self-organized, and
clustering and blocking are utilized to eliminate variability and bias from potential effect modifiers.
14. The system of claim 13 , further comprising a specific data inclusion window per independent variable and per cluster.
15. The system of claim 13 , wherein an incremental value of learning versus exploiting is continually assessed.
16. The system of claim 15 , wherein the continually assessing further comprises assessing a potential impact of adding, editing, interpolating, or removing independent variables.
17. The system of claim 1 , further comprising a continuous optimization module configured to identify, monitor, and improve clustering of experimental units process and explore/exploit management by further refining the effectiveness of recommended control elements and the system's hyperparameters.
18. The system of claim 1 , further comprising model-free operation wherein there is no need to know the underlying causal diagram or process mechanisms.
19. The system of claim 1 , further comprising model-based operation wherein causal learning provides a mathematical representation of physical connections and causal mechanisms in the system.
20. A computer-implemented method comprising:
injecting randomized controlled signals into the system of interest;
ensuring the signal injections occur within normal operational ranges and constraints;
monitoring at least one performance measure of the system of interest in response to the controlled signals;
computing confidence intervals about the causal effect of the presence relative to the absence of distinct signal injections on the performance measure of the system of interest;
selecting optimal signals for the performance of the system of interest based on the computed confidence intervals; and
computing confidence intervals by statistical testing on d-scores, which are defined as differences between measured effects when a variable is activated and when it is deactivated.