# US 2025/0149133 A1 · Adversarial imitation learning engine for kpi optimization

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

- Number: US 2025/0149133 A1
- Document ID: US20250149133A1
- Status: Published application
- Kind: A1
- Application number: 18/922,837
- Filed: 2024-10-22
- Published: 2025-05-08
- Earliest claimed: 2023-11-07
- Assignees: NEC Laboratories America, Inc.
- Applicants: NEC Laboratories America, Inc.
- Inventors: LuAn Tang; Yuyang Ye; Haifeng Chen; Haoyu Wang; Zhengzhang Chen; Wenchao Yu
- Web page: https://patentgrep.com/patent/US20250149133A1

## CPC

- G16H 10/60: for patient-specific data, e.g. for electronic patient records
- G06N 3/045: Combinations of networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/08: Learning methods
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N 3/094: Adversarial learning
- G06Q 10/0635: Risk analysis of enterprise or organisation activities
- G06Q 40/08: Insurance
- G16H 40/20: for the management or administration of healthcare resources or facilities, e.g. managing hospital staff or surgery rooms
- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert 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
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR

## Abstract

Systems and methods for optimizing key performance indicators (KPIs) using adversarial imitation deep learning include processing sensor data received from sensors to remove irrelevant data based on correlation to a final KPI and generating, using a policy generator network with a transformer-based architecture, an optimal sequence of actions based on the processed sensor data. A discriminator network is employed to differentiate between the generated action sequences and real-world high performance sequences employing. Final KPI results are estimated based on the generated action sequences using a performance prediction network. The generated action sequences are applied to the process to optimize the KPI in real-time.

## Claims

### Claim 1

A computer-implemented method for optimizing key performance indicators (KPIs) using adversarial imitation deep learning, comprising:
processing sensor data received from sensors to remove irrelevant data based on correlation to a final KPI;
generating, using a policy generator network with a transformer-based architecture, an optimal sequence of actions based on the sensor data;
differentiating between the optimal sequence of actions and real-world high performance sequences employing a discriminator network;
estimating final KPI results based on the optimal sequence of actions using a performance prediction network; and
applying the optimal sequence of actions to a process to optimize KPI in real-time.

### Claim 2 (depends on 1)

The method of claim 1, wherein the policy generator network employs a multi-head self-attention mechanism to capture temporal dependencies in the sensor data.

### Claim 3 (depends on 1)

The method of claim 1, wherein the discriminator network utilizes a neural network architecture to minimize discrepancies between generated action sequences and real-world high-performance sequences.

### Claim 4 (depends on 1)

The method of claim 1, wherein the performance prediction network employs a transformer-based architecture to estimate the final KPI results.

### Claim 5 (depends on 1)

The method of claim 1, further comprising training an environment simulator using variational autoencoder techniques to simulate future states of the process based on current actions and states.

### Claim 6 (depends on 5)

The method of claim 5, wherein the environment simulator is used to predict consequences of potential actions during generation of the optimal sequence of actions.

### Claim 7 (depends on 1)

The method of claim 1, further comprising selecting high KPI samples from historical data to train the policy generator network and the discriminator network.

### Claim 8 (depends on 1)

The method of claim 1, wherein the optimal sequence of actions to optimize the KPI include generated treatment action sequences for a patient's healthcare plan.

### Claim 9

A computer-implemented method for optimizing healthcare outcomes using adversarial imitation deep learning, comprising:
receiving patient data from one or more medical sensors monitoring a patient;
processing the patient data to remove irrelevant data based on correlation to a healthcare key performance indicator (KPI);
generating, using a policy generator network with a transformer-based architecture, an optimal sequence of treatment actions based on the patient data;
employing a discriminator network to differentiate between the optimal sequence of treatment actions and real-world high-performance treatment sequences;
estimating final healthcare KPI results based on the optimal sequence of treatment actions using a performance prediction network; and
applying the optimal sequence of treatment actions to a patient's care plan to optimize healthcare KPI in real-time.

### Claim 10 (depends on 9)

The method of claim 9, wherein the patient data includes real-time data and historical data.

### Claim 11 (depends on 9)

The method of claim 9, further comprising:
training an environment simulator using variational autoencoder techniques to simulate future patient states based on current treatment actions and patient states; and
using the environment simulator to predict consequences of potential treatment actions during generation of the optimal sequence of treatment actions.

### Claim 12 (depends on 9)

The method of claim 9, wherein the policy generator network employs a multi-head self-attention mechanism to capture temporal dependencies in the patient data.

### Claim 13

A system for optimizing healthcare outcomes using adversarial imitation deep learning, comprising:
a hardware processor; and
a memory storing instructions that, when executed by the hardware processor, cause the hardware processor to:
receive patient data from one or more medical sensors monitoring a patient;
process the patient data to remove irrelevant data based on correlation to a healthcare outcome metric;
generate, using a policy generator network with a transformer-based architecture, an optimal sequence of medical interventions based on the patient data;
employ a discriminator network to differentiate between the optimal sequence of medical interventions and real-world high-performance intervention sequences;
estimate healthcare outcome results based on the optimal sequence of medical interventions using a performance prediction network; and
apply action sequences to optimize the healthcare outcome metric in real-time.

### Claim 14 (depends on 13)

The system of claim 13, wherein the policy generator network employs a multi-head self-attention mechanism to capture temporal dependencies in the patient data.

### Claim 15 (depends on 13)

The system of claim 13, wherein the discriminator network utilizes a neural network architecture to minimize discrepancies between generated intervention sequences and real-world high-performance intervention sequences.

### Claim 16 (depends on 13)

The system of claim 13, wherein the performance prediction network employs a transformer-based architecture to estimate the healthcare outcome results.

### Claim 17 (depends on 13)

The system of claim 13, wherein the memory stores further instructions that, when executed by the hardware processor, cause the hardware processor to train an environment simulator using variational autoencoder techniques to simulate future patient states based on current interventions and patient states.

### Claim 18 (depends on 17)

The system of claim 17, wherein the environment simulator is used to predict consequences of potential medical interventions during generation of the optimal sequence of medical interventions.

### Claim 19 (depends on 13)

The system of claim 13, wherein the memory stores further instructions that, when executed by the hardware processor, cause the hardware processor to select high-performance healthcare outcome samples from historical patient data to train the policy generator network and the discriminator network.

### Claim 20 (depends on 14)

The system of claim 14, wherein the optimal sequence of medical interventions includes at least one of medication administration, surgical procedures, therapy sessions, and lifestyle recommendations.
