In-Database Machine Learning Model Serialization and Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current database systems lack efficient solutions for training a machine learning model in one instance and applying it in another, due to technical expertise requirements and inefficient resource consumption, particularly in network bandwidth and additional workload.
Innovation Solution
A system that automatically serializes a trained machine learning model into a binary file for easy transfer and recreation across different database instances, eliminating the need for retraining and ensuring model consistency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If users manually copy the trained machine learning model from one database instance to another, then the model can be transferred between instances, but it requires technical expertise and consumes network bandwidth
Solution Approach 1:
The patent implements automatic copying of trained machine learning models from a first database instance to a second database instance. The system serializes the trained model into a binary file format and automatically transfers it, eliminating the need for manual copying by users. This resolves the contradiction by maintaining model transfer capability while removing the technical expertise barrier.
Solution Approach 2:
The system performs self-service by automatically identifying when a model needs to be transferred between database instances and executing the transfer process without user intervention. The patent describes automatic model serialization and transfer mechanisms that operate autonomously, reducing the operational burden on users while maintaining adaptability across instances.
2Adaptability or versatility
If users manually copy the trained machine learning model between database instances, then model transfer is possible, but network bandwidth is consumed by data transfer
Solution Approach 1:
The patent serializes the trained machine learning model into a compact binary file format for efficient transfer. This serialization process optimizes the model representation to minimize data size and transfer requirements, enabling model copying between instances while reducing network bandwidth consumption compared to transferring raw model data or retraining datasets.
Solution Approach 2:
The system extracts only the essential model parameters and architecture information needed for model recreation, storing them in a serialized binary format. This extraction approach separates the critical model components from the full training dataset, allowing efficient transfer of only what is necessary for model deployment, thereby reducing network bandwidth usage.
3Adaptability or versatility
If the machine learning model is retrained in the second database instance after training in the first instance, then the model can be used locally, but additional workload is created affecting system speed and effectiveness
Solution Approach 1:
The patent implements preliminary action by training the machine learning model in advance in the first database instance and automatically transferring the trained model to the second instance before it is needed. This eliminates the need for retraining at the second instance, as the model is already prepared and available for immediate use, thereby maintaining productivity and avoiding additional workload.
Solution Approach 2:
The system copies the trained model from the first database instance to the second instance, allowing the second instance to use the pre-trained model directly. This copying approach is significantly more efficient than retraining, as it transfers only the model parameters and architecture rather than reprocessing the entire training dataset, thus preserving system productivity while ensuring model availability across instances.
Data Source
AI summary
In some example embodiments, a computer-implemented method may include training a machine learning model in a first database instance using a machine learning algorithm and a training dataset in response to receiving a request to train, serializing the trained machine learning model into a binary file in response to the training of the machine learning model, recreating the trained machine learning model in a second database instance using the binary file in response to receiving a request to apply the machine learning model, and generating an inference result by applying the recreated trained machine learning model on the inference dataset in the second database instance.