Model Sharing

Blackbox Sharing

Blackbox sharing is the distribution of a simulation run (or set of runs) to others without exposing the code, data or parameters that resulted in the model’s production, nor allowing any user-interaction with the model.

In the case of a simulation model built atop HASH, blackbox sharing allows:

• →

  simulations builders to share pre-computed runs of a simulation with others, without revealing the model’s workings;

• →

  simulation viewers to execute or inspect pre-computed runs of a simulation.

Blackbox sharing is useful for distributing forecasts and projections that show how systems evolve over time, without revealing source code or sensitive data. But blackbox sharing isn’t the only way to distribute a simulation run (at least in HASH).

Closed-Greybox Sharing

Closed-greybox sharing is similar to blackbox sharing, except that users are provided with the ability to vary a fixed set of parameters that feed a simulation (usually a simulation’s global variables).

As in blackbox sharing, simulation logic such as code and other parameters not intended for sharing can remain off-limits to simulation viewers.

Closed-greybox sharing is useful for enabling outsiders to vary key assumptions of a simulation model and conduct runs without revealing proprietary source code or sensitive data.

Open-Greybox Sharing

Open-greybox sharing goes one step further, allowing users to vary any input parameters in a simulation, or even connect their own data-sources for use.

At the same time, all or some model logic (simulation code) or sensitive data (e.g. personally-identifiable information) remains off-limits to viewers.

Given the ability of simulation viewers to vary parameters in an uncontrolled fashion it may, however, be possible to reverse-engineer simulation logic. For models of any reasonable degree of complexity this requires great effort, and is often infeasible – but sharers should be aware. Typically viewers of open-greybox models are asked to agree to some terms of engagement, or sign a non-disclosure/non-compete.

Open-greybox sharing is useful for sharing simulations with semi-trusted third-parties with whom a high degree of experimental freedom is to be entrusted.

Transparent Sharing

In the case of transparent sharing, access to all simulation source code and data is exposed to the end-user. This includes both:

1. →

   simulations that are shared completely (i.e. transparently) with one or more users, but kept private;

2. →

   simulations that are made public — for example published to hIndex, the largest repository of transparently shared simulations in the world.

When a simulation is shared in a transparent fashion, it may still be copyrighted, or subject to other license terms.

Output-Only Sharing

A final type of sharing, “output-only sharing”, is more restrictive than any of the other forms of sharing outlined above.

Output-only sharing refers to the distribution of certain specified simulation outputs only (i.e. single values of specific metrics tracked during a simulation run). The “entire run” of a simulation will not be visible, and only the individual metric tracked will be made available. Typically this is accessed via an Application Programming Interface, such as the HASH API.

Whereas blackbox sharing results in the distribution of an entire simulation run to the end-consumer, so they can see how it evolved over time, output-only sharing simply cherrypicks a metric (e.g. the final state of a particular value at the end of a simulation run, or the average value of a particular thing over the lifetime of a run).

Output-only sharing is useful for powering business intelligence dashboards, and applications which require operationalized simulation models to drive their own internal logic (e.g. a recommendation engine, or optimizer).

Getting started

HASH is the only AI and simulation platform that supports true blackbox, closed-greybox, open-greybox, transparent and output-only forms of sharing.

This makes HASH ideal for regulated environments in which both model secrecy and proving compliance is a concern, as well as high-security environments, and in powering third-party applications which require simulation data as inputs. Read more about sharing in HASH >

Currently available:

• →

  Transparent sharing is fully available within the hCore.

• →

  Closed-greybox sharing is in public beta and available when sharing or embedding private projects from hCore (with code-privacy guarantees coming soon).

• →

  Output-only sharing is available as part of HASH’s pre-release program (register for access).

• →

  Blackbox sharing and open-greybox sharing are coming soon.