Model Training

Model training is made possible by Obit FedDDL (Federated Distributed Deep Learning for the Web). Tensorflow.js models can be trained and distributed automatically in as little as 30 lines!

The following types of Workload are supported out of the box:

• MNist

• Mnist with differential privacy

• Mnist with weight compression

• Mnist (as an autoencoder/unsupervised learning task)

• Many Reinforcement Learning workloads are supported on their respective page

The following datasets are supported out of the box:

• MNist

• Reinforcement Learning simulations (see implemented simulations on the dedicated page)

Anyone writing a Workload is free to customize their model and dataset. For reinforcement learning, RL algorithms and environments may also be customized.

See the Algorithm and Environment APIs pages if the algorithms or environments provided are insufficient for your use case.

We also support Python environments to allow automatic support with many existing RL libraries, with a few major caveats. Learn more on the Python Environments page.