Composite ML Model for Ordered Code Sequence Generation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing techniques for generating ordered code sequences are inefficient, lack scalability, and generate inaccurate results due to their inability to effectively leverage multiple machine learning and/or rules-based models.

Innovation Solution

A composite machine learning model that jointly trains multiple different machine learning and/or rules-based models to generate an ordered code sequence, improving efficiency, accuracy, and scalability.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional single-model machine learning approaches are used to generate ordered code sequences, then the system is simpler to implement, but the accuracy and efficiency deteriorate

Engineering Contradiction:

Improveaccuracy of ordered code sequence generationVSAvoidcomplexity of machine learning model system

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent combines multiple machine learning models (including sequence models and graph models) into a composite model system. The sequence model generates unordered code sequences, which are then processed by the graph model to determine temporal relationships and produce the final ordered code sequence. This merging of multiple specialized models achieves higher accuracy than any single model could alone.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The code sequence generation task is divided into two distinct segments: first, the sequence model segment generates the unordered code sequence from input data; second, the graph model segment processes this unordered sequence to establish temporal relationships and generate the final ordered output. This segmentation allows each model to specialize in its specific function, improving overall system accuracy.

Inventive Principle:

Principle #1Segmentation

2Productivity

If conventional machine learning models are used for code sequence generation, then the computational resources required are lower, but the processing speed and efficiency deteriorate

Engineering Contradiction:

Improveefficiency of ordered code sequence generationVSAvoidcomputational resource consumption

Core Design Contradiction:

ProductivityVSUse of energy by moving object

Solution Approach 1:

The generation process is segmented into two efficient stages: the sequence model quickly generates unordered code sequences, and the graph model efficiently processes these sequences to establish temporal order. This segmentation allows computationally intensive operations to be distributed across specialized models, improving overall processing efficiency despite increased resource usage.

Inventive Principle:

Principle #1Segmentation

3Adaptability or versatility

If conventional models are used to generate ordered code sequences from large quantities of input data, then the system is easier to maintain, but the scalability deteriorates

Engineering Contradiction:

Improvescalability for large ordered sequences and input dataVSAvoidcomplexity of model architecture

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The composite model system is designed with universal components that can handle various types of input data and generate code sequences of different lengths and complexities. The sequence model and graph model work together as a universal framework that scales effectively from small to large datasets, maintaining consistent performance across different data volumes.

Inventive Principle:

Principle #6Universality (Multi-functionality)

4Measurement precision

If conventional single-model approaches are used, then the training process is simpler, but the generation accuracy deteriorates

Engineering Contradiction:

Improveaccuracy of code sequence predictionVSAvoidtraining time for composite model

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The training process is segmented into two phases corresponding to the two models: the sequence model is trained to generate unordered code sequences, and the graph model is trained to process these sequences and establish temporal relationships. This segmented training approach, while requiring more time than single-model training, achieves superior prediction accuracy by allowing each model to specialize in its specific task.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20250087351A1Ordered code sequences using a composite machine learning model

Publication Date: 2025.03.13 OPTUM INC

AI summary

Various embodiments of the present disclosure provide techniques for generating an ordered code sequence. For example, the techniques may include generating a predictive group code and an anchor code for an entity based on entity data. The techniques may include generating, using the first portion of the composite machine learning model, an unordered code sequence comprising one or more predicted codes based on the entity data, the predictive group code, and the anchor code. The techniques may include generating, using a second portion of the composite machine learning model, an ordered code sequence based on the unordered code sequence. The techniques may include providing the ordered code sequence.