Decomposing Tasks Through Artificial Intelligence Chaining

Link to the online patent PDF

Overview

This patent describes a framework for building complex, AI-driven services by breaking them down into smaller, manageable tasks. The core idea is to use an Intelligent Element Framework that manages and connects multiple, modular Intelligent Entities (IEs). Each IE is a specialized AI or machine learning model trained for a specific function. These IEs can be "chained" together, allowing the output of one model to be used as the input for another, creating a flexible and powerful system for automating information services.

The Problem

Building and maintaining large, monolithic AI systems is challenging. Such systems are often rigid, difficult to update, and not easily adaptable to new or evolving tasks. When a complex service requires multiple types of analysis or prediction (e.g., network fault detection, performance forecasting, and security analysis), creating a single, all-encompassing model is inefficient. This approach lacks modularity, making it difficult to reuse components, distribute workloads effectively, or update individual parts of the system without affecting the whole.

The Solution

The patent proposes a flexible architecture built around an Intelligent Element Framework that orchestrates modular AI components. The key elements are:

1. Intelligent Entities (IEs): These are self-contained, containerized AI/ML models, each trained to perform a highly specific task (e.g., predict traffic, analyze security threats, or manage capacity).
2. IE Catalog: A repository where all available IEs are stored and managed, making them discoverable and reusable.
3. Chaining Configuration: This defines how different IEs are connected in a sequence or a more complex directed graph. The framework uses this configuration to pass the output from one IE as an input to the next, effectively breaking a large problem into a series of smaller, specialized steps.
4. Orchestration: The framework integrates with Operational Support Systems (OSS) and orchestrators to deploy and manage these IE chains across a network of computing devices, from centralized cloud servers to edge devices.

Why It Matters

This architecture provides a blueprint for creating highly agile, scalable, and reusable AI-driven services. Its significance lies in:

• Modularity and Reusability: Instead of building new, complex models from scratch, developers can assemble new services by chaining together existing, proven IEs from a catalog.
• Scalability and Efficiency: Complex tasks are decomposed and can be distributed across a network, optimizing resource usage and improving performance.
• Agility and Maintainability: Individual IEs can be updated, improved, or replaced without disrupting the entire system. This allows for continuous improvement and rapid adaptation to new requirements.
• Enabling Sophisticated AI Services: This model makes it practical to build complex, multi-step AI applications for telecommunications and other industries, such as fully automated network management, predictive maintenance, and dynamic resource allocation.

Relevance Beyond Telecommunications

The principle of decomposing complex problems by chaining modular AI components is a powerful paradigm with applications far beyond telecommunications:

• Manufacturing and Industrial IoT: In a smart factory, a process can be modeled as a chain of IEs. For example, an initial IE could analyze sensor data to predict wear and tear on a machine, its output feeding into a second IE that schedules maintenance, which in turn triggers a third IE to re-route production to other machines.
• Healthcare and Genomics: Complex diagnostic workflows can be built by chaining specialized models. A chain could consist of one IE that analyzes medical images (like an MRI), a second that processes genetic markers from a blood sample, and a third that integrates these outputs to assess a patient's risk for a specific disease.
• Financial Services: A sophisticated fraud detection system could be built by chaining IEs. One IE might analyze transaction patterns, another could verify user location data, and a third could cross-reference the activity with historical behavior. The combined, chained analysis would provide a much more robust and accurate fraud assessment than any single model could.
• Autonomous Systems: The decision-making process for an autonomous vehicle can be broken down into a chain: a perception IE identifies objects, its output feeds a prediction IE that forecasts their movement, and a final planning IE uses that information to decide on the vehicle's path.

This approach allows for the creation of highly sophisticated, yet manageable and adaptable, AI systems in virtually any domain that deals with complex, multi-step processes.

Technical Details

The system is composed of several key architectural components:

• Intelligent Element Framework (110): The central component responsible for managing the lifecycle of IEs, processing data through various channels, and executing the chaining configuration.
• Intelligent Entities (120): Modular, specialized AI/ML models. The patent provides several examples, including:
  • Fault IE: For detecting and managing system faults.
  • Capacity IE: For monitoring and managing resource capacity.
  • Performance IE: For evaluating system performance metrics.
  • Security IE: For managing security threats.
  • Alarm Reduction IE and Alarm RCA IE: For reducing false alarms and performing root cause analysis.
  • Traffic Forecasting IE: For predicting network traffic patterns.
• Chaining Configuration (606): A definition that specifies the sequence and connections between IEs, forming a processing pipeline.
• IE Catalog (115): A repository that stores and manages the available IEs, making them accessible to the framework.
• Integration with Orchestration: The framework is designed to work alongside an OSS Management Orchestration (410) layer, which includes orchestrators, network managers, and other operational support systems to manage the underlying infrastructure.

Status: Issued

Application Number: 16/121,494

Patent Number: 11960976

Filing Date: 2018-09-04

Issue Date: 2024-04-16

‍