18340732. TECHNIQUES FOR KNOWLEDGE DISTILLATION BASED MULTI-VENDOR SPLIT LEARNING FOR CROSS-NODE MACHINE LEARNING simplified abstract (QUALCOMM Incorporated)
TECHNIQUES FOR KNOWLEDGE DISTILLATION BASED MULTI-VENDOR SPLIT LEARNING FOR CROSS-NODE MACHINE LEARNING
Organization Name
Inventor(s)
June Namgoong of San Diego CA (US)
Yiyue Chen of San Diego CA (US)
Taesang Yoo of San Diego CA (US)
Abdelrahman Mohamed Ahmed Mohamed Ibrahim of San Diego CA (US)
TECHNIQUES FOR KNOWLEDGE DISTILLATION BASED MULTI-VENDOR SPLIT LEARNING FOR CROSS-NODE MACHINE LEARNING - A simplified explanation of the abstract
This abstract first appeared for US patent application 18340732 titled 'TECHNIQUES FOR KNOWLEDGE DISTILLATION BASED MULTI-VENDOR SPLIT LEARNING FOR CROSS-NODE MACHINE LEARNING
Simplified Explanation
The techniques described in this patent application involve using a machine learning algorithm to train encoders from multiple UE vendors and a shared decoder from a gNB vendor to create a universal gNB decoder capable of decoding input from UEs from different vendors.
- Machine learning algorithm used to train encoders from multiple UE vendors and a shared decoder from a gNB vendor
- Development of a universal gNB decoder capable of decoding input from UEs from different vendors
- Comparable performance and overhead to specific decoders developed for each encoder
---
- Potential Applications
- Mobile communication networks
- Internet of Things (IoT) devices
- Smart city infrastructure
- Problems Solved
- Compatibility issues between UEs from different vendors and gNB decoders
- Overhead of developing specific decoders for each encoder
- Performance variations between different decoders
- Benefits
- Improved interoperability between UEs and gNB decoders
- Simplified decoder development process
- Enhanced performance and efficiency in decoding input from various UEs
Original Abstract Submitted
The techniques described herein utilize a machine learning algorithm to train the encoders from multiple UE vendors and a shared decoder from a gNB vendor in order to develop a universal gNB decoder that may be capable of decoding input from UEs from different UE vendors at comparable performance and overhead to different decoders that are specifically developed for each encoder.
