DeRelayL: A Sustainable Decentralized Relay Learning Framework
This paper introduces DeRelayL, a novel training paradigm for sustainable decentralized relay learning. It addresses the high financial and computational barriers of large-scale model training, which currently exclude common users like mobile data creators. Existing collaborative methods, particularly federated learning, focus on data privacy and group aggregation but lack sustainability and user ownership. DeRelayL proposes a new approach to enable common users to train and share models collaboratively.
Key facts
- Large-scale machine learning models require high financial and computational resources.
- Only technological giants and well-funded institutions can afford such training.
- Common users, especially mobile users, are excluded from benefiting due to barriers.
- Current methods for accessing large-scale models limit user ownership or lack sustainability.
- There is a need for a collaborative model training approach for common users.
- Existing collaborative paradigms like federated learning focus on data privacy and group-based model aggregation.
- DeRelayL is proposed as a novel training paradigm to address these issues.
- The paper is published on arXiv with ID 2605.02935.
Entities
Institutions
- arXiv