Incident Remediation With Pre-Trained Models and Few-Shot Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning proposals for incident identification in industrial settings require extensive training epochs, leading to prolonged resource utilization and reduced hardware lifespan, hindering timely response to incidents.
Innovation Solution
Employ pre-trained machine learning models with limited further training, utilizing a foundation model architecture and specific task models trained with few-shot learning to efficiently identify and remediate incidents, reducing resource consumption and response time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If extensive training epochs are used for machine learning models, then incident identification accuracy is improved, but resource utilization time increases and hardware lifespan decreases
Solution Approach 1:
The patent applies preliminary action by pre-training foundation models before they are needed for incident identification. The models are trained in advance on historical incident data and stored for later use. When an incident occurs, the pre-trained model can immediately process the incident without requiring extensive training at the time of detection, thus improving identification accuracy while avoiding prolonged resource utilization that would reduce hardware lifespan.
Solution Approach 2:
The patent uses copying by creating a pre-trained model that replicates the knowledge and patterns learned from extensive training data. Instead of performing extensive training epochs at the time of incident detection, the system copies the learned patterns into a pre-trained model that can quickly and accurately identify incidents, thereby maintaining high accuracy while reducing the time hardware needs to be actively trained.
2Measurement precision
If extensive training epochs are used for machine learning models, then incident identification accuracy is improved, but resource utilization time increases
Solution Approach 1:
The system performs preliminary training of foundation models in advance, storing the trained models for rapid deployment. When an incident occurs, the pre-trained model can immediately process the incident data without requiring time-consuming training epochs, thus achieving high identification accuracy while minimizing response time loss.
Solution Approach 2:
The patent copies learned patterns and knowledge into pre-trained models that are ready for immediate use. This copying approach allows the system to maintain high incident identification accuracy while avoiding the time loss associated with performing extensive training epochs at the moment of incident detection.
3Use of energy by moving object
If pre-trained models with limited training are used, then resource consumption is reduced, but incident identification accuracy may decrease
Solution Approach 1:
The patent segments the training process into two phases: an initial phase where foundation models are trained on comprehensive historical incident data, and a subsequent phase where the pre-trained models are deployed for incident identification. This segmentation allows the system to invest computing resources upfront during training, then minimize resource consumption during operation while maintaining high accuracy through the pre-trained models' learned patterns.
Solution Approach 2:
The system performs preliminary training action during the foundation model development phase, where extensive computing resources are allocated to learn from historical incident data. Once the models are pre-trained, the system can reduce computing resource consumption during incident detection while maintaining high identification accuracy, as the models already contain the necessary learned patterns and knowledge.
Data Source
AI summary
Methods, computer program products, and systems are presented. The method computer program products, and systems can include, for instance: evaluating alert data received from one or more computer environment in reference to a criterion; detecting that a current incident has occurred based on the criterion being satisfied; performing similarity analysis between the current incident and one or more historical incident; identifying, from the similarity analysis, a match between the current incident and the one or more historical incident; responsively to the identifying of the match, training a predictive model for production of a trained predictive model with use of dataset data of the one or more historical incident and historical text based data describing the one or more historical incident, wherein the historical text based data has been defined by an administrative user.