On-Demand Training Dataset Requests to Reduce Air-Interface Overhead
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The continuous transmission of training datasets by a network device to a training device during AI model training results in resource waste and overheads, as the training device does not always need all datasets.
Innovation Solution
The training device requests specific training datasets based on related information, such as dataset size, configuration, and performance metrics, allowing the network device to send only necessary datasets, thereby optimizing resource use.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the network device continuously transmits training datasets to the training device, then the training device can have sufficient training data available, but air interface resources and overheads are wasted
Solution Approach 1:
The training device autonomously determines its own training data needs by evaluating model performance metrics and training progress, then requests only the necessary datasets from the network device. This self-service mechanism eliminates continuous transmission of unnecessary data, reducing air interface resource waste while ensuring sufficient training data availability.
Solution Approach 2:
The system transitions from static continuous transmission to dynamic on-demand transmission. The training device dynamically adjusts its data requests based on real-time training state, performance metrics, and convergence assessment, allowing the network device to transmit datasets only when and what is actually needed.
2Quantity of substance
If the network device sends training datasets periodically, then data availability is maintained, but unnecessary datasets are transmitted causing overhead increase
Solution Approach 1:
The training device continuously monitors training performance metrics and model convergence status, providing feedback to the network device about actual data needs. This feedback loop enables the network device to send datasets selectively based on real training requirements rather than following a fixed periodic schedule, reducing overhead while maintaining availability.
Solution Approach 2:
The training device performs preliminary assessment of its training needs by evaluating current performance metrics and convergence criteria before requesting datasets. This preliminary action allows the device to precisely determine what data is needed next, preventing overhead associated with transmitting datasets that would not be used.
3Productivity
If the training device requests specific datasets based on needs, then resource utilization improves, but communication overhead for requests increases
Solution Approach 1:
The system changes the parameter of data transmission from continuous/periodic time-based scheduling to need-based event-triggered scheduling. By changing the triggering parameter from time to training state conditions (performance metrics, convergence status), the system achieves higher resource utilization while the request overhead is minimized through efficient condition-based triggering.
Data Source
AI summary
This application provides a training dataset obtaining method and an apparatus. A training device may request a network to send a training dataset, and request information also indicates related information of a first training dataset that is sent by a network device to the training device and that is needed by the training device. In other words, in this application, the training device may indicate a needed training dataset to the network device, and the network device may send, to the training device, a training dataset indicated by the training device, and does not need to continuously deliver training datasets. The method can reduce a waste of air interface resources and air interface overheads, and improve use performance of the air interface resources.