Pluralis Research is developing Protocol Learning, an approach to collaborative model training that distributes computational workloads among multiple participants while preventing any single entity from accessing the complete model.
This method allows participants to contribute computational resources without requiring direct ownership or control over the full model. It aims to facilitate participation without upfront costs and establish a sustainable economic structure for contributors. A key challenge in this approach is achieving low-bandwidth, model-parallel training, a problem widely regarded as unsolved. Pluralis Research is working on solutions to this issue.
