Self-supervised learning with swarm intelligence refers to a learning paradigm in which multiple autonomous learning agents interact in a decentralized network to learn robust representations through mutual consensus building and collective pseudo-label generation.
Formally, a swarm consists of a set of agents \( A = \{a_1, a_2, ..., a_n\} \), each possessing local models \( M_i \) and exchanging information via communication protocols \( C \).
The agents generate self-supervised signals \( S_i \) from their local data and share these with other agents to define collective training objectives that improve learning performance.
This distributed learning architecture promotes diversity of representations and increases robustness against failures and noise.
Definition:
“Self-supervised learning with swarm intelligence is a decentralized learning method in which multiple autonomous agents jointly learn meaningful data representations without external labels through interaction and collective pseudo-label generation.”
Source:
Li, X., & Wang, Y. (2022). Swarm Intelligence Enhanced Self-Supervised Learning: A Decentralized Approach. Proceedings of the 36th AAAI Conference on Artificial Intelligence.