PROJECT EINSTEIN IS UNDER CONSTRUCTION AND NOT USABLE AT THIS POINT. THIS PAGE WILL BE UPDATED AS FUNCTIONALITY BECOMES AVAILABLE.
Einstein is a secure execution environment for conducting machine learning trials against confidential data.
The goal of Einstein is to enable collaboration between data scientists and organizations with interesting problems. The challenge is that interesting problems come with interesting data sets that are almost always proprietary. These data sets are rich with trade secrets and personably identifiable information, and are usually encumbered by contracts, regulated by statute, and subject to corporate data stewardship policies.
In-house data science departments know how to work with this data, but the compliance issues make it is hard for them to collaborate with third parties and experts from industry and academia.
Einstein aims to solve this problem by creating a sandbox for machine learning experiments inside the environment that hosts sensitive data. With Einstein, a company can host machine learning challenges and invite third parties to submit solutions for evaluation against sensitive data that would otherwise be unavailable.
For more information, please see the Einstein Design Overview, which defines terminology, goals, design facets, and a Russian doll model for feature delivery.
The Living Specification shows proposed workflows, based on a command-line tool.
The Architectural Design page contains links to other design documents.
Einstein is currently in the design phase. We have built a mockup of the Einstein services. You can see an interactive session in our living spec.
- TODO: instructions for building and deploying Einstein.
- TODO: sample benchmark and candidate tutorial
- TODO: Documentation Homepage
- CLI reference
- TODO: Einstein concepts
- TODO: Creating a benchmark
- TODO: Creating a candidate
- TODO: Submitting a candidate
- TODO: Analyzing test runs
- TODO: Sample Benchmark and Candidate
- Architectural Design Homepage
- TODO: Overview
- TODO: Goals and Priciples
- TODO: Concepts
- TODO: Labratory Service
- TODO: CLI
- TODO: Benchmark
- TODO: Candidate
- TODO: Test Run Repository
- TODO: Test Run Analysis
- TODO: contribution guidelines
- TODO: source tree layout
- TODO: coding guidelines