AI/ML Learning Block Training for Low-Overhead Wireless Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training artificial intelligence or machine learning models in communication networks are inefficient and lack flexibility in adapting to different computing capabilities and resource availability of user equipment and base stations.
Innovation Solution
A method for training an AI/ML model involves receiving training configuration information for a learning block of less than all layers of the model, determining and training the block using user equipment or base station resources, and transmitting parameters associated with the block, allowing for flexible adaptation of forward-propagation-only and backpropagation training methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the entire AI/ML model is trained using traditional methods, then the model achieves comprehensive learning, but the training overhead and signaling overhead increase significantly
Solution Approach 1:
The patent segments the AI/ML model into multiple learning blocks, where each block contains a subset of layers that can be trained independently. This segmentation allows the system to train only necessary portions of the model rather than the entire model, thereby reducing training overhead and signaling overhead while maintaining essential learning functionality.
2Reliability
If traditional training methods are used, then complete model training is achieved, but flexibility in adapting to different computing capabilities is reduced
Solution Approach 1:
The patent implements dynamic adaptability by allowing the selection and configuration of different learning blocks based on the computing capabilities and resource availability of the user equipment and base station. The system can dynamically adjust which layers are trained together in each learning block, making the training process flexible and adaptable to varying computational environments.
3Ease of operation
If all layers are trained together in one block, then training is simplified, but the ability to perform incremental learning and reduce overhead is lost
Solution Approach 1:
The patent divides the model into multiple learning blocks with different granularities, allowing the system to select appropriate block sizes based on specific training needs. This segmentation enables incremental learning where only necessary blocks are trained, improving training efficiency and reducing overhead while maintaining operational simplicity through configurable block structures.
4Loss of time
If incremental learning with learning blocks is implemented, then training overhead is reduced, but the complexity of managing multiple blocks and configurations increases
Solution Approach 1:
The patent creates a universal learning block structure that can serve multiple functions and be configured in various ways. The same learning block framework can accommodate different numbers of layers, different training configurations, and different computing capabilities, reducing the need for separate management mechanisms for each scenario and thereby reducing overall system complexity.
Data Source
AI summary
Aspects of the present disclosure provide methods and apparatuses for training an artificial intelligence or machine learning (AI/ML) model to support deep neural network (DNN)-based applications and DNN-based services in a communication network. According to some embodiments, a user equipment (UE) may receive, from a base station (BS), training configuration information for a learning block comprising one or more successive layers of the AI/ML model. The learning block may include a subset of less than all layers of the AI/ML model. The UE may determine the learning block using the training configuration information. The UE may train the AI/ML model or the learning block using the training configuration information. The UE may transmit, to the BS, one or more parameters associated with the learning block.