# US 2025/0095814 A1 · Determining and performing optimal actions on a system

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0095814 A1
- Document ID: US20250095814A1
- Status: Published application
- Kind: A1
- Application number: 18/537,794
- Filed: 2023-12-12
- Published: 2025-03-20
- Earliest claimed: 2023-09-20
- Assignees: Microsoft Technology Licensing, LLC
- Applicants: Microsoft Technology Licensing, LLC
- Inventors: Chao MA; Jiaqi ZHANG; Cheng ZHANG; Joel JENNINGS; Colleen TYLER; Marife DEFANTE; Lisa Lynne PARKS
- Web page: https://patentgrep.com/patent/US20250095814A1

## CPC

- G16H 20/00: ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
- G06N 3/045: Combinations of networks
- G06N 5/046: Forward inferencing; Production systems
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

In certain examples, a causal inference model is trained on a re-balancing task in a self-supervised manner, using ‘unlabelled’ training data pertaining to multiple domains. Rather than approaching casual inference as a domain-specific task (e.g., designing one causal-inference approach for a particular manufacturing application, another for a particular aerospace application, another for a specific medical application etc.,) a general-purpose causal inference mechanism is learned from a large, diverse training set that contains many treatments dataset over many field/applications (e.g., combining manufacturing data, engineering data, medical data etc. in a single dataset used to train a single neural network). In other words, a cross-domain causal inference model is trained, which can then be applied to a dataset in any domain, including domains that were not explicitly encountered by the neural network during training.

## Claims

### Claim 1

A computer-implemented method, comprising:
receiving a first training dataset specific to a first domain, the first training dataset comprising a first covariate matrix characterizing a first system and a first treatment vector encoding a first treatment observation relating to the first system;
receiving a second training dataset specific to a second domain, the second training dataset comprising a second covariate matrix characterizing a second system and a second treatment vector encoding a second treatment observation relating to the second system;
computing using a causal inference model applied to the first training dataset a first forward mode output corresponding to the first treatment vector;
computing using the causal inference model applied to the second training dataset a second forward mode output corresponding to the second treatment vector;
training the causal inference model based on a training loss that quantifies error between:
the first treatment vector and the first forward mode output, and
the second treatment vector and the second forward mode output, resulting in a trained causal inference model:
computing a rebalancing weight vector using the trained causal inference model applied to a third dataset specific to a third domain, the third dataset comprising a third covariate matrix characterizing a third system, a third treatment vector encoding a third treatment observation and a third outcome vector;
estimating based on the third outcome vector and the rebalancing weight vector a causal effect associated with the third treatment vector;
based on the causal effect, determining a treatment action; and
performing the treatment action on at least one target system belonging to the third domain.

### Claim 2 (depends on 1)

The method of claim 1, wherein the third dataset is specific to a third domain, wherein the causal inference model is not exposed to any data from the third domain during training.

### Claim 3 (depends on 1)

The method of claim 1, wherein the first training dataset, the second training dataset and the third dataset are each non-randomized.

### Claim 4 (depends on 1)

The method of claim 1, wherein the at least one third system comprises the at least one target system.

### Claim 5 (depends on 1)

The method of claim 1, wherein the at least one target system comprises a machine and the causal effect comprises an estimated treatment effect pertaining to performance of the machine.

### Claim 6 (depends on 5)

The method of claim 5, wherein the machine is a manufacturing machine, and the estimated treatment effect pertains to:
quality of a product manufactured using the machine, or
production efficiency of the machine.

### Claim 7 (depends on 1)

The method of claim 1, wherein the at least one target system comprises a computer system and the causal effect comprises an estimated treatment effect pertaining to usage of memory or processing resources.

### Claim 8 (depends on 1)

The method of claim 1, wherein the causal inference model generates during training:
a first output value, wherein the forward mode output corresponding to the first training dataset is computed based on the first output value and a first normalization factor computed from the first covariate matrix, and
a second output value, wherein the forward mode output corresponding to the second training dataset is computed based on the second output value and a second normalization factor computed from the second covariate matrix;
wherein the rebalancing weight vector is computed based on: a third output value computed by the trained causal inference model, the third treatment vector, and a third renormalization factor computed from the third covariate matrix.

### Claim 9 (depends on 1)

The method of claim 1, wherein the causal effect is determined based on a summation of a product of: the rebalancing weight vector, the third treatment vector and the third outcome vector.

### Claim 10 (depends on 1)

The method of claim 1, wherein the causal inference model has a transformer neural network architecture.

### Claim 11

A computer system comprising:
at least one memory configured to store computer-readable instructions; and
at least one hardware processor coupled to the at least one memory, wherein the computer-readable instructions are configured to cause the at least one hardware processor to implement operations comprising:
receiving a first training dataset specific to a first domain, the first training dataset comprising a first covariate matrix characterizing a first system and a first treatment vector encoding a first treatment observation relating to the first system;
receiving a second training dataset specific to a second domain, the second training dataset comprising a second covariate matrix characterizing a second system and a second treatment vector encoding a second treatment observation relating to the second system;
computing using a causal inference model applied to the first training dataset a first forward mode output corresponding to the first treatment vector;
computing using the causal inference model applied to the second training dataset a second forward mode output corresponding to the second treatment vector;
training the causal inference model based on a training loss that quantifies error between:
the first treatment vector and the first forward mode output, and
the second treatment vector and the second forward mode output, resulting in a trained causal inference model:
computing a rebalancing weight vector using the trained causal inference model applied to a third dataset specific to a third domain, the third dataset comprising a third covariate matrix characterizing a third system, a third treatment vector encoding a third treatment observation relating to the third system and a third outcome vector;
estimating based on the third outcome vector and the rebalancing weight vector a causal effect associated with the third treatment vector;
based on the causal effect, determining a treatment action.

### Claim 12 (depends on 11)

The computer system of claim 11, wherein said operations comprise:
automatically performing the treatment action on at least one target system belonging to the third domain.

### Claim 13 (depends on 11)

The computer system of claim 11, wherein the at least one third system comprises the at least one target system.

### Claim 14 (depends on 11)

The computer system of claim 11, wherein the third dataset is specific to a third domain, wherein the causal inference model is not exposed to any data from the third domain during training.

### Claim 15 (depends on 11)

The computer system of claim 11, wherein the first training dataset, the second training dataset and the third dataset are each non-randomized.

### Claim 16 (depends on 11)

The computer system of claim 11, wherein the causal effect comprises an estimated treatment effect pertaining to performance of a machine.

### Claim 17 (depends on 16)

The computer system of claim 16, wherein the machine is a manufacturing machine, and the estimated treatment effect pertains to:
quality of a product manufactured using the machine, or
production efficiency of the machine.

### Claim 18 (depends on 16)

The computer system of claim 16, wherein the causal effect comprises an estimated treatment effect pertaining to usage of memory or processing resources by a computer system.

### Claim 19 (depends on 11)

The computer system of claim 11, wherein the causal inference model has a transformer neural network architecture.

### Claim 20

Computer-readable storage media embodying computer readable instructions, the computer-readable instructions configured upon execution on at least one hardware processor to cause the at least one hardware processor to implement operations comprising:
computing a rebalancing weight vector using a trained causal inference model applied to a third dataset specific to a third domain, the trained causal inference model having been trained by:
receiving a first training dataset specific to a first domain, the first training dataset comprising a first covariate matrix characterizing a first system and a first treatment vector encoding a first treatment observation relating to the first system,
receiving a second training dataset specific to a second domain, the second training dataset comprising a second covariate matrix characterizing a second system and a second treatment vector encoding a second treatment observation relating to the second system, and
computing using a causal inference model applied to the first training dataset a first forward mode output corresponding to the first treatment vector;
computing using the causal inference model applied to the second training dataset a second forward mode output corresponding to the second treatment vector;
training the causal inference model based on a training loss that quantifies error between:
the first treatment vector and the first forward mode output, and
the second treatment vector and the second forward mode output,
resulting in a trained causal inference model;
the third dataset comprising a third covariate matrix characterizing a third system, a third treatment vector encoding a third treatment observation of the third system and a third outcome vector;
estimating based on the third outcome vector and the rebalancing weight vector a causal effect associated with the third treatment vector;
based on the causal effect, determining a treatment action.
