Wireless AI/ML Training Data Selection Using Data State Information
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI/ML model training processes in wireless communication systems face challenges in determining data importance, especially in 6G networks, leading to increased transmission overhead and latency, which degrades overall performance due to unclear data importance determination for data of the same type and reliance on higher-layer QoS indicators.
Innovation Solution
Implement methods and devices that utilize data state information (DSI) to selectively transmit AI/ML model training data based on DSI thresholds and report formats, reducing unnecessary data transmission and enhancing performance by minimizing signaling overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all AI/ML model training data is transmitted to ensure complete model training, then model training completeness is improved, but transmission overhead and latency increase significantly
Solution Approach 1:
The patent applies local quality by determining data state information (DSI) for individual data samples to identify which specific samples are important for model training. Instead of treating all data uniformly, the system evaluates each sample's contribution to model convergence and selectively transmits only those samples with high DSI values, thereby reducing transmission overhead while maintaining training effectiveness.
Solution Approach 2:
The patent implements partial action by transmitting only a subset of training data samples that meet the DSI threshold criteria. Rather than transmitting all available data (excessive action), the system identifies and transmits only the necessary portion of data that contributes meaningfully to model training, reducing unnecessary transmission overhead and latency.
2Loss of energy
If data state information (DSI) calculation and selective transmission is implemented, then transmission overhead is reduced, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by calculating data state information (DSI) for training data samples before transmission decisions are made. The system pre-evaluates each data sample's importance metric and compares it against a threshold in advance, so that only samples meeting the criteria are selected for transmission. This preliminary filtering simplifies the overall system complexity by making transmission decisions based on pre-computed metrics rather than complex real-time evaluations.
3Loss of time
If DSI threshold filtering is applied to select important data, then transmission latency is reduced, but model training accuracy may deteriorate
Solution Approach 1:
The patent applies parameter changes by adjusting the DSI threshold parameter to control the balance between transmission latency and model training accuracy. The system can dynamically modify the threshold value based on network conditions, device capabilities, and training progress. By changing this key parameter, the system optimizes the trade-off: higher thresholds reduce transmission overhead and latency, while lower thresholds ensure more comprehensive data transmission for improved training accuracy.
Data Source
AI summary
Aspects of the present disclosure provide methods and devices for artificial intelligence or machine learning (AI/ML) model training in a wireless communication network. A first device receives, from a second device, AI/ML model training assistance information and information related to transmission of respective AI/ML model training data, collectively or separately. The first device determines data state information (DSI) of the respective AI/ML model training data based on the AI/ML model training assistance information. The first device transmits, to the second device, the respective AI/ML model training data based on at least one of the DSI of the respective AI/ML model training data or the information related to the transmission of respective AI/ML model training data.