Contrastive Learning for Anomalous Event Prediction Without Labels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine-learning techniques struggle to accurately predict anomalous events due to the lack of comprehensive training labels, leading to sub-optimal prediction accuracy.
Innovation Solution
Utilize contrastive learning to train encoders for unstructured and contextual data, generating feature vectors in separate and shared feature spaces, and combine them to enhance predictiveness, without the need for labeled training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing machine-learning techniques are used with training labels, then the model can be trained, but prediction accuracy is sub-optimal due to incomplete training labels for various anomalous event types
Solution Approach 1:
The system uses self-supervised learning where the model learns from unstructured data and contextual data without requiring external labels. The contrastive learning mechanism enables the model to self-improve by finding similarities and differences between data representations, allowing it to predict anomalous events without complete training labels.
Solution Approach 2:
The patent introduces an intermediary representation layer that transforms both unstructured data and contextual data into a shared feature space. This intermediary space allows the model to compare and contrast different data types, enabling accurate anomaly detection even without explicit training labels for all event types.
2Measurement precision
If multiple data sources are processed separately, then each data type can be analyzed independently, but prediction effectiveness is reduced due to lack of integrated information
Solution Approach 1:
The patent merges unstructured data and contextual data into a unified representation by transforming both into the same feature space. This combining allows the model to leverage information from multiple sources simultaneously, improving prediction effectiveness while managing complexity through a standardized transformation process.
Solution Approach 2:
The patent creates a universal feature space that can accommodate multiple data types (unstructured data and contextual data). This multi-functional approach allows the same processing mechanism to handle different data sources, reducing overall system complexity while improving integrated analysis capability.
Data Source
AI summary
Various techniques can include systems and methods for using contrastive learning to predict anomalous events in data processing systems. The method can include accessing an unstructured data file and contextual data associated with the unstructured data file. The method can also include generating an event-data input element for the unstructured data file. The event-data input element can include a set of feature vectors. The set of feature vectors can include a first feature vector generated by using a first encoder to process the unstructured file and a second feature vector generated by using a second encoder to process the contextual data. The method can also include generating a classification result of the unstructured data file by using a machine-learning model to process the event-data input element, in which the classification result includes a prediction of whether the particular event corresponds to an anomalous event.