World Scale
Distributed Computing Platform

It is a distributed computing network where excess computing resources are connected peer-to-peer and used as if they were a single GPU machine.
The ecosystem is designed by managing computing resource providers and users are connected world wide.

It is a distributed computing network where excess computing resources are connected peer-to-peer and used as if they were a single GPU machine.
The ecosystem is designed by managing computing resource providers and users are connected world wide.

The Structure of Emeth

Data Parallelism

Data Parallel is a method in which the machine learning training material data is subdivided and assigned to each node, and the results computed by each are aggregated from time to time.
The intermediate machine learning model obtained during the computation process is shared among each node and the learning process proceeds in a coordinated manner. This method is the simplest and most effective distributed computation method.

Model Parallelism

Model parallel copies the entire training material data to each node, with each node taking on a certain step of the training process.
This method requires more inter-node communication, but the resulting machine learning model is more accurate.

Hybrid Parallelism

The final point of our roadmap is a hybrid parallel that combines the above two methods.
By subdividing the learning material while simultaneously distributing the processing steps, efficient use of computational resources becomes possible.
The final aggregated machine learning model is the optimal model that balances processing time and result accuracy.

Our Team

Team

Yoshikazu Nishimura

CTO / Blockchain Specialist

Team

Takuya Obata

Blockchain Specialist

Team

Satoshi Hirano

Deep Learning Researcher

Team

Vialli Muhammad Faijar

Deep Learning Researcher

Materials

Icon
GitHub
The source code for the client required to provide computing power to Emeth is available here. Once installed and configured on each participant's machine, the client will automatically connect to the Emeth network and start providing computing power.
Icon
Docs
Information on the opening phase of the "Alt Emeth" distributed computing infrastructure.
This phase is the next stage of the test phase and is positioned as a semi-open test phase just prior to the go-live phase. We are inviting applications from the general public, so please participate.