RESPECT AI™

Advancing responsible AI adoption

More

View All

In Numbers

92%

According to the survey, conducted on behalf of RBC by Maru/Matchbox, companies currently using AI/analytics agree it is important for businesses to implement AI in an ethical way. However, 92 per cent have concerns in dealing with the ethical challenges that AI represents, and just over half have someone responsible for ethical development of data and AI technology.

88%

The results of the survey also highlighted some significant challenges that businesses face in terms of bias such as race and gender. The vast majority (88 per cent) of companies believe they have bias within their organization, but almost half of them do not understand the challenges that bias presents in AI.

1 /

RESPECT AI™ Pillars

  • An icon of a cube.

    The ability of an AI system to defend against adversarial attacks. This component of RESPECT AI™ includes Advertorch, Borealis AI’s well-established adversarial robustness research code, which implements a series of attack and defense strategies that can be used to protect against risks. This tool is offered to AI researchers and scientists that aim to advance the field of robustness in machine learning.

Tutorials & Research

View All

1 /

Open Source Tools

AdverTorch

This toolbox provides machine learning practitioners with the ability to generate private and synthetic data samples from real world data. It currently implements 5 state of the art generative models that can generate differentially private synthetic data.

GitHub

LiteTracer

LiteTracer acts as a drop-in replacement for argparse, and it can generate unique identifiers for experiments in addition to what argparse already does. Along with a reverse lookup tool, LiteTracer can trace-back the state of a project that generated any result tagged by the identifier.

GitHub

Private Synthetic Data Generation

This toolbox provides machine learning practitioners with the ability to generate private and synthetic data samples from real world data.It currently implements 5 state of the art generative models that can generate differentially private synthetic data.

GitHub