The Department of Defense often works in austere environments with low or limited bandwidths that severely restrict the ability to conduct machine learning analysis.
The Technica Decentralized Federated Learning System (TDFLS) looks to change this paradigm. Leveraging a new framework based on the InterPlanetary File System (IPFS), TDFLS restructured the federated learning data sharing structure to minimize bandwidth needs so it can work in edge environments.
As further outlined in a new Technica Innovation Solution paper, TDFLS allows many mobile nodes to collaborate in training machine learning models without relying on a central entity. Instead, each node only needs to send a fraction of the data used in traditional centralized federated learning frameworks.
By leveraging TDFLS, military organizations can more reliably use tactical edge computing solutions in Denied-Disrupted, Intermittent, and Limited (D-DIL) environments. TDFLS will allow military organizations to use IFPS and a private network to initiate a machine learning process or join one already in progress.
Early results show that TDFLS scales with the number of participants, continues to work with intermittent connectivity, requires minimal resources, and guarantees the accuracy of the trained model within 1% of a traditional centralized federated learning framework.
In using TDFLS, the military can:
- Adapt current and future machine learning algorithms in a DFL model
- Develop new algorithms to be used in DFL environments
- Improve system resilience for D-DIL and contested environments, or if assets become lost
- Access DFL across heterogeneous network environments and platforms
- Enhance communication efficiency
Visit technicacorp.com and follow us on LinkedIn, Twitter, and Facebook to read about more cutting-edge technology solutions we are developing and delivering to our customers.
Read more about the Technica Decentralized
Federated Learning System:
Come join Technica and put your skills to work in defending our national security!