Safety Net Control for ML Predictive Routing Behavior

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Network administrators are hesitant to fully automate routing decisions using machine learning-based predictive engines due to the risk of unpredictable network performance and potential disruptions, as existing systems lack a mechanism to ensure acceptable performance and manage risk levels effectively.

Innovation Solution

The safety net adjusts the operation of predictive routing engines, using machine learning to detect detrimental actions and adjust behavior based on predefined policies, allowing for a tradeoff between risk-taking and conservative approaches to manage automation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If machine learning-based predictive routing engine is used to automate routing decisions, then network automation and proactive failure detection are improved, but network performance stability and predictability deteriorate

Engineering Contradiction:

Improverouting automationVSAvoidnetwork performance stability

Core Design Contradiction:

Extent of automationVSReliability

Solution Approach 1:

The safety net engine continuously monitors the predictive routing engine's decisions and compares them against actual network outcomes. When detrimental actions are detected (such as unnecessary reroutes or failures to reroute when needed), the safety net engine feeds back adjustments to modify the predictive engine's behavior, creating a closed-loop system that maintains automation while ensuring performance stability.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The safety net engine acts as an intermediary layer between the predictive routing engine and the actual network routing decisions. It intercepts decisions from the predictive engine, evaluates them against learned patterns of detrimental behavior, and either approves or blocks them. This mediator role allows the system to maintain high automation while filtering out unreliable decisions.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If the predictive routing engine takes aggressive actions to prevent failures, then proactive failure detection is improved, but network disruptions increase due to unnecessary rerouting

Engineering Contradiction:

Improvefailure detection capabilityVSAvoidnetwork disruptions

Core Design Contradiction:

ReliabilityVSObject-generated harmful factors

Solution Approach 1:

The safety net engine learns from historical data what constitutes detrimental actions in advance, building a model of harmful patterns before they occur. This preliminary learning allows the system to predict when the predictive routing engine is likely to make unnecessary rerouting decisions and prevent them before they cause network disruptions, while still allowing beneficial proactive failure detection.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system dynamically adjusts the operational parameters of the predictive routing engine based on learned patterns. When the safety net engine detects that the predictive engine is becoming too aggressive, it modifies parameters such as the threshold for triggering reroutes or the confidence level required before acting. This parameter adjustment allows the system to maintain effective failure detection while reducing harmful disruptions.

Inventive Principle:

Principle #35Parameter changes

3Reliability

If manual control is maintained over routing decisions, then network performance stability is preserved, but productivity and response time to failures deteriorate

Engineering Contradiction:

Improveperformance stabilityVSAvoidfailure response efficiency

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

Instead of fully automating routing decisions, the system implements partial automation where the safety net engine selectively overrides the predictive routing engine only when detrimental actions are detected. This partial automation approach maintains human-level performance stability while achieving machine-level response speed for the majority of decisions, avoiding the need for complete manual control.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20260025327A1Safety net engine for machine learning-based network automation

Publication Date: 2026.01.22 CISCO TECHNOLOGY INC

AI summary

In one embodiment, a device obtains data regarding routing decisions made by a machine learning-based predictive routing engine for a network. The device determines, based on the data regarding the routing decisions, a behavior of the machine learning-based predictive routing engine. The device compares the behavior of the machine learning-based predictive routing engine to a behavioral policy for the machine learning-based predictive routing engine. The device adjusts operation of the machine learning-based predictive routing engine, when the behavior of the machine learning-based predictive routing engine violates the behavioral policy.