Memri allows data owners (like you!) to provide access to build models on their data without sharing that data. Privacy-protected ML pipelines offer the ability to quickly and easily train ML models with real world nlp, computer vision and structured data.
By submitting this form, I acknowledge receipt of the Memri’s Privacy Policy.
The ecosystem that allows you to build machine learning Memri ecosystem consists of 3 main components:
A flutter web client that enables you to view and interact with your data using pluggable data apps.
written in Rust with a SQLite database providing a Graph API to store and access your data
Python plugins that import your data, or infer new data by applying machine learning on your data.
Our remote machine-learning architecture is based on the open-source OpenMined codebase. Using PySyft Plans, you can define machine learning pipelines that can be executed remotely to learn from data without seeing that data. The basic idea is to send a pipeline with a model over the network to a data owner, send the data through the pipeline, and download the updated model back.
@make_plan
def pipeline(dl = dataloader, model = model):
optimizer = remote_torch.optim.SGD(model.parameters()) x, y = next(dl)
out = model(x=x)
loss = remote_torch.nn.functional.cross_entropy(out, y)
loss.backward()
optimizer.step()
return model
We continuously generate data and information about ourselves but we have little control or access to it. Many of the services that we use prioritise data collection, while compromising our freedoms, agency and trust.
At Memri, we are building an open community and tackling these challenges through shared creativity and collaboration. We build tools that empower people and help create the new web - one that is private, trustworthy and free from centralized control.
By submitting this form, I acknowledge receipt of the Memri’s Privacy Policy.
