Self-Intelligent Improvement in Predictive Data Models

Link to the online patent PDF

‍

Overview

The patent for "Self-Intelligent Improvement in Predictive Data Models" introduces a system that automates the enhancement of machine learning models. It describes a "model assessor" that continuously generates and evaluates new "candidate models" by combining inputs and outputs from existing models. If a new candidate model proves to be more accurate by showing a better correlation with actual outcomes, it automatically replaces the older model. This creates a dynamic ecosystem where predictive models can evolve and improve over time without direct human intervention.

‍

The Problem

Traditional predictive computer models, such as those used for forecasting network congestion, weather patterns, or financial trends, are static and require manual updates by data scientists or engineers. This manual process creates significant bottlenecks, is slow, and relies heavily on human domain knowledge that is difficult to program into a computer. The core problem is the absence of an automated, intelligent mechanism to continuously maintain and improve the accuracy of these predictive models as new data becomes available.

‍

The Solution

The invention proposes a closed-loop system that enables models to improve themselves. The key components are:

1. Self-Intelligent Entities (SIEs): These are individual, validated predictive models (e.g., regression models) that operate within a containerized environment (like Docker). Each SIE generates predictions based on specific data inputs.
2. Model Assessor: This is the brain of the system. It retrieves predicted outputs from various SIEs and generates new candidate models. A candidate model is created by augmenting an existing model's inputs with the predictive outputs from other, different models.
3. Correlation Testing & Automated Updates: The model assessor calculates a correlation score between the candidate model's predictions and the actual, real-world outcomes. If the score exceeds a predefined threshold—meaning the new model is more accurate—it automatically replaces the original SIE. If not, the candidate model is discarded. This creates a continuous cycle of testing and improvement.

Why It Matters

This patent is significant because it provides a blueprint for creating predictive systems that are not static but are instead self-adapting and self-improving. This has profound implications for any industry relying on predictive analytics, including telecommunications, finance, healthcare, and logistics. By automating the model improvement process, this system makes predictive analytics more scalable, efficient, and resilient, reducing the need for constant human oversight and allowing models to adapt to changing conditions in real-time.

‍

Relevance Beyond Telecommunications

The concept of self-improving predictive models is not limited to telecommunications and has transformative potential across numerous sectors:

• Financial Services: Algorithmic trading models could use this system to continuously adapt to changing market conditions. A model assessor could test new trading strategies (candidate models) in real-time and automatically deploy those that demonstrate higher profitability or lower risk, leading to more resilient and intelligent trading systems.
• Healthcare: In medical diagnostics, predictive models that analyze patient data (e.g., medical images, lab results) to detect diseases could be continuously improved. As new patient cases and outcomes become available, the system could refine its diagnostic accuracy, potentially leading to earlier and more reliable detection of illnesses.
• E-commerce and Retail: Recommendation engines that predict customer purchasing behavior can become more accurate. The system could test new recommendation algorithms based on emerging trends and automatically implement those that lead to higher engagement and sales.
• Energy and Utilities: Predictive maintenance models for power grids or machinery can be made more effective. By analyzing sensor data, the system could continuously refine its ability to predict equipment failures, allowing for more efficient maintenance scheduling and reducing downtime.

This architecture provides a universal framework for creating dynamic, learning systems that improve their own performance over time, making it a foundational technology for the next wave of AI applications.

‍

Technical Details

The system is designed with a microservices-style architecture:

• Regression Techniques: The initial SIEs can be built using a variety of methods, including linear, logistic, polynomial, and ridge regression.
• Containerization: SIEs are deployed on a container platform like Docker or Kubernetes, ensuring they are isolated, scalable, and easy to manage.
• Asynchronous Communication: A topic queue (e.g., Kafka, RabbitMQ) is used for communication. SIEs publish their predictions to topics, and the model assessor subscribes to these topics to gather data for generating candidate models.
• Model Assessor Components: The assessor includes a data cache, a timestamp matcher to synchronize data inputs and outputs, and a correlation tester to evaluate model performance.
• Iterative Loop: The entire process is an iterative loop of generating, testing, and either replacing or discarding models, as illustrated here.

Status: Issued

Application Number: 16/261,176

Patent Number: 11410063

Filing Date: 2019-01-29

Issue Date: 2022-08-09

‍