ML Classifier for Unclassified Software Discovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The manual discovery and mapping of configuration items in computer networks is time-consuming, and existing automated processes face delays and inefficiencies when dealing with numerous and frequently updated software applications, leading to temporary classification failures.
Innovation Solution
The implementation of machine learning classifiers as a backup or redundant system to classify software applications based on characteristic files, which can quickly provide classifications without additional delay, even when discovery patterns fail, and can be used in conjunction with rule-based discovery patterns for redundancy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated software processes are used to discover and map configuration items, then productivity is improved, but device complexity increases and the system requires periodic updates to maintain reliability
Solution Approach 1:
The system segments the discovery process into distinct components: discovery patterns for structured detection, machine learning classifiers for categorization, and modular software agents. Each component handles specific tasks independently, improving maintainability and reducing overall system complexity while maintaining high productivity
Solution Approach 2:
Discovery patterns are pre-configured with detection rules, file path templates, and classification criteria before deployment. This preliminary preparation enables the automated processes to immediately begin discovery operations without requiring complex real-time decision-making, thereby improving productivity while keeping the runtime system simpler
2Measurement precision
If discovery patterns are updated to match new software applications, then measurement precision is improved, but loss of time increases due to the update process
Solution Approach 1:
The system enables self-service through machine learning classifiers that automatically learn and adapt to new application types without requiring manual pattern updates. The ML models are trained on characteristic files and application behaviors, automatically generating classification rules that maintain measurement precision while eliminating the time loss associated with manual pattern development
Solution Approach 2:
Instead of structurally modifying discovery patterns to accommodate new applications, the system changes parameters by training ML classifiers with new data samples. This parameter-based adaptation approach maintains the existing pattern framework while improving classification accuracy for new applications, significantly reducing update time
3Reliability
If machine learning classifiers are used as a backup to discovery patterns, then reliability is improved, but device complexity increases
Solution Approach 1:
The system introduces an intermediary layer between discovery patterns and classification output: machine learning classifiers serve as mediators that receive discovery data and produce classifications. This intermediary approach improves reliability by providing alternative classification paths while managing complexity through a clear layered architecture where each layer has defined responsibilities
4Ease of operation
If manual discovery and mapping is performed, then ease of operation is maintained, but productivity decreases significantly
Solution Approach 1:
The automated discovery system is designed with universal components that can handle multiple application types and scenarios through configurable patterns and trained ML models. This multi-functionality allows the system to perform complex automated discovery tasks while maintaining operational flexibility similar to manual processes, achieving high productivity without sacrificing ease of operation
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computing system includes a database and a discovery application that initiates discovery of software applications within a managed network by probing the computing devices for applications in stalled thereon. The discovery application executes discovery patterns in an attempt to classify a discovered application and determines that the discovered application is unclassified after executing the discovery patterns. In response to determining that the discovered application is unclassified, the discovery application (i) generates a configuration item representing the discovered application without classification and (ii) selects a characteristic file containing attributes of the discovered application. The discovery application provides the characteristic file to a machine learning (ML) classifier trained using a plurality of characteristic files. The discovery application receives, from the ML classifier, a classification of the discovered application, modifies the configuration item to include the classification of the discovered application, and stores, in the database, the configuration item as modified.