Deep Learning Data Labeling via Dual Scoring Endpoint Preprocessing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep-learning models face challenges in accurately evaluating and labeling feedback data due to the absence of pre-processing logic in traditional methods, leading to inefficient resource allocation and difficulty in determining the relationship between training data and transformative algorithms.
Innovation Solution
A method involving the deployment of two scoring endpoints, where the second endpoint preprocesses native data and outputs it along with a user-generated score to the first endpoint, enabling the creation of a comprehensive data set for accurate evaluation and retraining of the deep-learning model by matching raw and transformed payloads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used to evaluate and label feedback data for deep-learning models, then the process is simpler to implement, but the accuracy and efficiency of model retraining deteriorates due to missing pre-processing logic
Solution Approach 1:
The patent introduces an intermediary component that captures and stores the pre-processing logic and transformation parameters as feedback data. This intermediary layer bridges the gap between the simple traditional evaluation method and the complex pre-processing requirements, allowing accurate model retraining without directly exposing the complexity of pre-processing logic to the evaluation process.
Solution Approach 2:
The patent implements a feedback mechanism where transformation parameters and pre-processing logic are captured, stored, and fed back into the model training process. This feedback loop enables the model to be retrained with accurate transformation information, improving evaluation precision while maintaining a manageable system structure through automated feedback collection.
2Adaptability or versatility
If pre-processing logic is embedded in the deep-learning model, then the model can process native data directly, but it becomes difficult to extract and define the transformation logic for feedback data labeling
Solution Approach 1:
The patent extracts the pre-processing logic and transformation parameters from the embedded model context and captures them as separate feedback data. By taking out these transformation details, the system maintains the model's ability to process native data while making the transformation logic accessible for feedback data labeling and analysis.
Solution Approach 2:
The patent creates a copy of the transformation parameters and pre-processing logic as feedback data alongside the model's native data processing capability. This copying approach allows the system to maintain versatility in processing native data while having accessible copies of transformation logic for evaluation and retraining purposes.
3Loss of information
If resources are dedicated to determine the relationship between training data and transformative algorithms, then the transformation logic can be understood, but the resource consumption becomes burdensome
Solution Approach 1:
The patent implements a self-service mechanism where the system automatically captures and stores transformation parameters and pre-processing logic as feedback data during normal operation. This eliminates the need for burdensome manual resource dedication to understand transformation relationships, as the system self-documented the transformations through automated feedback collection.
Solution Approach 2:
The patent performs preliminary capture and storage of transformation parameters before they are needed for model retraining. By pre-capturing the transformation logic as feedback data during normal model operation, the system avoids the need for resource-intensive analysis when retraining is required, significantly reducing energy and computational consumption.
Data Source
AI summary
A first and second scoring endpoint with payload logging are deployed. At the second scoring endpoint, native data and a user-generated score for the native data are received, the native data is pre-processed into readable data for the deep-learning model, and the user-generated score and the readable data are output to the first scoring endpoint, which is associated directly with the deep-learning model. A raw payload that includes the native data is output to a payload store. At the first scoring endpoint, the readable data and the user-generated score are processed by the deep-learning model, which outputs a transformed payload and a prediction, respectively, to the payload store. The raw payload is matched with the transformed payload and the prediction to produce a comprehensive data set, which is evaluated to describe a set of transformation parameters. The deep-learning model is retrained to account for the set of transformation parameters.