Wireless AI Training Data Signaling With Augmentation Tags
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing complexity of network planning and resource scheduling in wireless communication networks due to diverse service demands and advanced technologies poses challenges, particularly in implementing artificial intelligence for efficient network operation and energy saving.
Innovation Solution
A communication method and apparatus that utilizes data augmentation techniques to enhance model performance by distinguishing between first-type and second-type training data, allowing flexible and efficient processing of AI models in wireless communication networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data augmentation is performed to obtain second-type training data from first-type training data, then model performance is improved, but signaling overhead increases due to need to indicate data types
Solution Approach 1:
The patent merges the data type indication with the training data transmission by embedding type identifiers within the data structure itself. Multiple groups of training data are transmitted together with a single set of identification information that indicates the types of all groups, thereby reducing the signaling overhead compared to transmitting separate indications for each data group.
Solution Approach 2:
The training data is segmented into multiple groups, where each group can be of first-type or second-type data. The identification information is structured to correspond to these segments, allowing efficient indication of data types across multiple groups with reduced overhead compared to individual indications for each data element.
2Measurement precision
If multiple groups of training data are transmitted with individual identification information for each group, then data type tracking is precise, but signaling overhead increases
Solution Approach 1:
The patent combines the identification information for multiple data groups into a unified structure. A single set of identification information indicates the types of multiple groups of training data simultaneously, maintaining precise tracking of data types while significantly reducing the total signaling overhead compared to individual indications for each group.
3Ease of operation
If only first-type training data is used for model processing, then data processing is simple, but model performance is limited
Solution Approach 1:
The patent implements a dynamic data processing approach where the system can flexibly select between first-type training data and second-type training data (or both) based on the specific processing task requirements. This dynamic selection mechanism maintains operational simplicity while improving model performance by using augmented data when beneficial.
4Adaptability or versatility
If network supports diverse services and advanced technologies, then network capability is enhanced, but network planning and resource scheduling complexity increases
Solution Approach 1:
The patent applies parameter changes by introducing identification information parameters that indicate the types of training data. This parameter-based approach allows the network to efficiently manage diverse services and advanced technologies without significantly increasing planning complexity, as the type indications enable systematic resource allocation and scheduling.
Data Source
AI summary
This disclosure provides a communication method and apparatus. The method includes: receiving a training data set and first information, where the first information includes identification information corresponding to one or more pieces of training data in the training data set, the identification information indicates that the corresponding training data belongs to first-type training data or second-type training data, and the second-type training data is obtained by processing the first-type training data based on an augmentation algorithm. In the method, different communication scenarios can be flexibly matched, and a model is processed based on the first-type training data and/or the second-type training data, thereby improving performance of the model.