Computational Storage Neural Model Mediation for Data Compatibility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of smooth data communication between multiple neural network models in computational storage systems, where each model operates using different data representations, leading to inefficiencies in data transmission and reception.
Innovation Solution
A computational storage system incorporating an interpreter neural network model that transforms data between neural network models using training and transformation tables, enabling seamless data exchange by mapping data to common parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple neural network models operate using different data representations, then each model can be optimized for its specific function, but communication between models becomes inefficient and complex
Solution Approach 1:
The patent introduces an interpreter neural network model as an intermediary that receives data from one neural network model, transforms it through learned mappings, and outputs data in the format required by another neural network model. This mediator resolves the communication incompatibility between models with different data representations without requiring changes to the models themselves, thus maintaining model-specific optimization while simplifying inter-model communication.
Solution Approach 2:
The interpreter neural network dynamically transforms data parameters between different neural network models by learning optimal mapping relationships. The system changes the parameter representations (data formats, feature encodings, normalization schemes) of incoming data to match the requirements of target models, enabling efficient communication while preserving the specialized functionality of each model.
2Reliability
If data transformation between neural network models is performed manually, then communication can be established, but the process requires significant time and computational resources
Solution Approach 1:
The system performs preliminary training of the interpreter neural network to pre-learn the data transformation mappings between different neural network models. During this offline training phase, the interpreter learns optimal transformation strategies by analyzing data from source and target models. Once trained, the interpreter can rapidly transform data during runtime without requiring complex manual transformation processes, significantly reducing transformation time while maintaining reliability.
Solution Approach 2:
The interpreter neural network autonomously performs data transformation without requiring manual intervention or complex external processing. The trained interpreter self-adapts to transform data between different model formats by applying learned mappings, eliminating the need for time-consuming manual transformation processes while ensuring consistent and reliable communication between models.
3Speed
If direct data transmission is used between neural network models, then speed is maintained, but data compatibility issues arise due to different representations
Solution Approach 1:
The interpreter neural network serves as a mediation layer that enables fast data transmission between neural network models with different data representations. By preprocessing data through the trained interpreter, the system maintains high transmission speeds while ensuring compatibility between models. The interpreter quickly transforms data formats without requiring slow manual intervention, thus preserving speed while solving compatibility issues.
Data Source
AI summary
A computational storage system includes: a storage device configured to store a first neural network model, a second neural network model, and an interpreter neural network model; and a computing device configured to transform data, transmitted and received between the first neural network model and the second neural network model, into a parameter corresponding to the data by using the interpreter neural network model, based on a first training table, a second training table, and a first transformation table, and transmit and/or receive the parameter between the first neural network model and the second neural network model.