Controlling a manufacturing process using causal models
An example method includes selecting treatment assignments for operation of a system in an environment based on a causal model mapped to a probability distribution over possible treatment assignments, where the causal model measures causal relationships between the treatment assignments and a performance metric associated with the environment. The method further includes mapping causal effects between the treatment assignments and measures of uncertainty associated with the environment in the causal model to corresponding probabilities using probability matching. The method further includes determining the performance metric using the treatment assignments and corresponding dependent variables obtained from the environment in response to the treatment assignments. The method further includes adjusting, based on the performance metric, the causal model, and re-computing the adjusted causal model by computing overall causal effects based on the measures of uncertainty around the overall causal effects.
1 . A method comprising:
clustering procedural instances based on one or more characteristics of an environment to form two or more clusters;
identifying each of the procedural instances as one of a hybrid instance or a baseline instance for a controllable element associated with the environment;
selecting first control settings for the identified baseline instances;
selecting second control settings for the identified hybrid instances;
selecting treatment assignments for operation of a system in the environment based on a causal model mapped to a probability distribution over possible treatment assignments, wherein the causal model measures causal relationships between the treatment assignments and a performance metric associated with the environment, selection of the treatment assignments being for blocked groups according to a blocking scheme, the blocked groups formed based on a first blocked group including the baseline instances and a second blocked group including the hybrid instances;
mapping causal effects between the treatment assignments and measures of uncertainty associated with the environment in the causal model to corresponding probabilities using probability matching;
determining the performance metric using the treatment assignments and corresponding dependent variables obtained from the environment in response to the treatment assignments;
adjusting, based on the performance metric, the causal model; and
re-computing the adjusted causal model by computing overall causal effects based on the measures of uncertainty around the overall causal effects.
2 . The method of claim 1 , wherein the treatment assignments are selected for blocked groups according to the blocking scheme by selecting control settings for controllable elements for entities in the environment.
3 . The method of claim 1 , further comprising:
selecting a spatial extent for each respective entity of a plurality of entities associated with the environment;
selecting a temporal extent for each respective procedural instance of the procedural instances; and
generating the procedural instances based on the selected spatial extents and the selected temporal extents.
4 . The method of claim 1 , wherein clustering the procedural instances comprises:
computing d-scores for the first and second control settings, wherein each respective d-score represents an effect of the respective treatment assignment being active;
monitoring interaction terms between the causal effects and one or more external factors associated with the environment; and
applying a factorial analysis of variance (ANOVA) to generate the two or more clusters across space and time such that each respective cluster of the two or more clusters is representative of a distinct state of the one or more external factors.
5 . The method of claim 1 , wherein clustering the procedural instances comprises applying a machine learning technique.
6 . The method of claim 5 , wherein the machine learning technique comprises one of a decision tree-based technique, a conditional inference tree-based technique, a deep neural network, or a k-means clustering technique.
7 . A system comprising:
a memory configured to store data associated with an environment; and
a processor in communication with the memory, the processor being configured to:
cluster procedural instances based on one or more characteristics of the environment to form two or more clusters;
identify each of the procedural instances as one of a hybrid instance or a baseline instance for a controllable element associated with the environment;
select first control settings for the identified baseline instances;
select second control settings for the identified hybrid instances;
select treatment assignments for operation of a system in the environment based on a causal model mapped to a probability distribution over possible treatment assignments, wherein the causal model measures causal relationships between the treatment assignments and a performance metric associated with the environment, selection of the treatment assignments being for blocked groups according to a blocking scheme, the blocked groups formed based on a first blocked group including the baseline instances and a second blocked group including the hybrid instances;
map causal effects between the treatment assignments and measures of uncertainty associated with the environment in the causal model to corresponding probabilities using probability matching;
determine the performance metric using the treatment assignments and corresponding dependent variables obtained from the environment in response to the treatment assignments;
adjust, based on the performance metric, the causal model; and
re-compute the adjusted causal model by computing overall causal effects based on the measures of uncertainty around the overall causal effects.
8 . The system of claim 7 , wherein the treatment assignments are selected for blocked groups according to the blocking scheme by selecting control settings for controllable elements for entities in the environment.
9 . The system of claim 7 , wherein the processor is further configured to:
select a spatial extent for each respective entity of a plurality of entities associated with the environment;
select a temporal extent for each respective procedural instance of the procedural instances; and
generate the procedural instances based on the selected spatial extents and the selected temporal extents.
10 . The system of claim 7 , wherein to cluster the procedural instances, the processor is configured to:
compute d-scores for the first and second control settings, wherein each respective d-score represents an effect of the respective treatment assignment being active;
monitor interaction terms between the causal effects and one or more external factors associated with the environment; and
apply a factorial analysis of variance (ANOVA) to generate the two or more clusters across space and time such that each respective cluster of the two or more clusters is representative of a distinct state of the one or more external factors.
11 . The system of claim 7 , wherein to cluster the procedural instances, the processor is configured to apply a machine learning technique.
12 . The system of claim 11 , wherein the machine learning technique comprises one of a decision tree-based technique, a conditional inference tree-based technique, a deep neural network, or a k-means clustering technique.
13 . A non-transitory computer-readable medium encoded with instructions that, when executed, cause a processor to:
cluster procedural instances based on one or more characteristics of an environment to form two or more clusters;
identify each of the procedural instances as one of a hybrid instance or a baseline instance for a controllable element associated with the environment;
select first control settings for the identified baseline instances;
select second control settings for the identified hybrid instances;
select treatment assignments for operation of a system in the environment based on a causal model mapped to a probability distribution over possible treatment assignments, wherein the causal model measures causal relationships between the treatment assignments and a performance metric associated with the environment, selection of the treatment assignments being for blocked groups according to a blocking scheme, the blocked groups formed based on a first blocked group including the baseline instances and a second blocked group including the hybrid instances;
map causal effects between the treatment assignments and measures of uncertainty associated with the environment in the causal model to corresponding probabilities using probability matching;
determine the performance metric using the treatment assignments and corresponding dependent variables obtained from the environment in response to the treatment assignments;
adjust, based on the performance metric, the causal model; and
re-compute the adjusted causal model by computing overall causal effects based on the measures of uncertainty around the overall causal effects.
14 . The non-transitory computer-readable medium of claim 13 , wherein the treatment assignments are selected for blocked groups according to the blocking scheme by selecting control settings for controllable elements for entities in the environment.
15 . The non-transitory computer-readable medium of claim 13 , the non-transitory computer-readable medium being further encoded with instructions that, when executed, cause the processor to:
select a spatial extent for each respective entity of a plurality of entities associated with the environment;
select a temporal extent for each respective procedural instance of the procedural instances; and
generate the procedural instances based on the selected spatial extents and the selected temporal extents.
16 . The non-transitory computer-readable medium of claim 13 , wherein to cluster the procedural instances, the processor is configured to:
compute d-scores for the first and second control settings, wherein each respective d-score represents an effect of the respective treatment assignment being active;
monitor interaction terms between the causal effects and one or more external factors associated with the environment; and
apply a factorial analysis of variance (ANOVA) to generate the two or more clusters across space and time such that each respective cluster of the two or more clusters is representative of a distinct state of the one or more external factors.
17 . The non-transitory computer-readable medium of claim 13 , wherein to cluster the procedural instances, the processor is configured to apply a machine learning technique.
18 . The non-transitory computer-readable medium of claim 17 , wherein the machine learning technique comprises one of a decision tree-based technique, a conditional inference tree-based technique, a deep neural network, or a k-means clustering technique.