Federated, Split, or Split-Federated Learning for Network Intelligence in 6G? a Comprehensive Investigation of Deployment Challenges

The fastest development and the advancement towards the 6 G network strongly demand network intelligence applying $\text{AI} / \text{ML}$-driven approaches. 6 G is strongly driven by the accessibility of massive data and the shift from centralized and massive data to distributed and small data. This inclination has enthused researchers to adopt collaborative and distributed machine learning in 6G to improve and automate different tasks of telecommunication networks. Adopting distributed ML-powered approaches in the 6 G will play a noteworthy role. In contrast, their counterparts of centralized solutions will face severe challenges of communication along with computing resource constraints, overhead in computing, and privacy concerns. Federated Learning (FL), Split Learning (SL), and Split-Federated Learning (SFL) integrated intelligence in the 6 G network will be promising solutions. In this paper, the author investigates the compatible qualitative attributes of FL, SL, and SFL and their comparative deployment challenges considering the feasibility of lower model training time and privacy in the 6 G network, followed by recent developments and trends. It is observed that SFL will be a potential solution for 6 G network intelligence with faster and more secure model training.