Cooperative Semantic Knowledge Base Update Policy for Multiple Semantic Communication Pairs

Shuling Li,Yaping Sun,Jinbei Zhang,Kechao Cai,Hao Chen,Shuguang Cui,Xiaodong Xu

2024-10-03

Abstract:Semantic communication has emerged as a promising communication paradigm and there have been extensive research focusing on its applications in the increasingly prevalent multi-user scenarios. However, the knowledge discrepancy among multiple users may lead to considerable disparities in their performance. To address this challenge, this paper proposes a novel multi-pair cooperative semantic knowledge base (SKB) update policy. Specifically, for each pair endowed with SKB-enabled semantic communication, its well-understood knowledge in the local SKB is selected out and uploaded to the server to establish a global SKB, via a score-based knowledge selection scheme. The knowledge selection scheme achieves a balance between the uplink transmission overhead and the completeness of the global SKB. Then, with the assistance of the global SKB, each pair's local SKB is refined and their performance is improved. Numerical results show that the proposed cooperative SKB update policy obtains significant performance gains with minimal transmission overhead, especially for the initially poor-performing pairs.

Information Theory

What problem does this paper attempt to address?

This paper attempts to solve the performance inconsistency problem caused by knowledge differences among users in multi - user semantic communication scenarios. Specifically, each semantic communication pair (SemCom pair) has its own local semantic knowledge base (SKB), but due to different initial knowledge levels, some communication pairs may perform well while others perform poorly. This performance difference limits the efficiency of the overall system and the quality of user experience (QoE). To this end, the paper proposes a multi - pair cooperative semantic knowledge base update policy, aiming to optimize the global SKB by sharing the advantageous knowledge of each communication pair, and then improve the performance of all communication pairs, especially those with poor initial performance.

### Main contributions of the paper:
1. **Proposing a multi - pair cooperative SKB update policy**: This policy selects well - understood knowledge from different local SKBs for sharing to update the SKB. Experimental results show that for data initially mis - understood, the performance is significantly improved after cooperatively updating the SKB.
2. **Designing a score - based knowledge selection scheme**: This scheme is used to select uploaded knowledge and can flexibly adjust the uplink transmission overhead. Each communication pair can adjust its uplink transmission overhead through a pre - configured threshold, so as to share comprehensive knowledge while minimizing the transmission overhead.
3. **Numerical results verify the effectiveness of the proposed policy**: Experimental results show that the performance of communication pairs after cooperation is significantly improved, especially in communication pairs with poor initial performance.

### System model:
- **Semantic knowledge base (SKB)**: Each communication pair maintains an SKB, representing its knowledge level. The SKB is constructed by class - level attribute vectors, and each attribute vector is associated with a specific category, reflecting the state of a specific attribute.
- **SKB - supported semantic communication model**: This model supports image classification and image generation tasks. At the sending end, the semantic encoder extracts the semantic information of the input image and sends its index to the receiving end, thereby reducing the transmission load.

### Cooperative SKB update strategy:
1. **Update of local SKB**: Each communication pair updates its local SKB according to the test data after training the local model.
2. **Cooperative update process**: Each communication pair calculates the F1 score of each category and selects the advantageous knowledge to upload to the server through the score - based knowledge selection scheme. The server aggregates this knowledge to form a global SKB and broadcasts it to all communication pairs.
3. **Knowledge selection scheme**: By setting a threshold γ, the amount of uploaded knowledge is controlled, thereby balancing the uplink transmission overhead and the integrity of the global SKB.

### Numerical results:
- **Relationship between the number of uploaded attribute vectors and the threshold**: As the threshold γ increases, the number of uploaded attribute vectors decreases and the transmission load is reduced.
- **Relationship between transmission delay and the threshold**: As the threshold γ increases, the transmission delay is reduced.
- **Performance evaluation**: The performance of image classification and generation tasks is evaluated by F1 score and FID score. Experimental results show that the performance of communication pairs after cooperation is significantly improved, especially in communication pairs with poor initial performance.

### Conclusion:
The multi - pair cooperative SKB update policy proposed in the paper effectively solves the knowledge difference problem in multi - user semantic communication. By sharing advantageous knowledge, the performance of the overall system is improved, especially for communication pairs with poor initial performance.