Empowering a new era of AI development.
Explore the technology behind Hybrid Intelligence to understand the advantages of Neuro-symbolic AI.
TECHNOLOGY
A framework for developing fully-fledged neuro-symbolic AI solutions.
Available as an easy-to-use developer library designed to make AI simple to create and easy to integrate.
HYBRID INTELLIGENCE
easier than you think
-
Induce
a Hybrid Intelligence model directly from data in a single run.
-
Query
the model to get predictions together with precise explanations in real-time.
-
Decide
build fit-for-purpose decisions from predictions and explanations.
INDUCTION
Induction is the Hybrid Intelligence training method that fits a neuro-symbolic AI model from training data.
Model training is orderly and methodical, producing the best possible model in a single training run. It is a self-contained process, accessed via a single API call that gives the developer control over model structure and optimisation objectives.
The outcomes are observable and measurable, minimising resource usage and building stakeholder confidence and project predictability.
Performance
Generalises better than deep learning techniques, making it more data-efficient and better equipped to handle data outliers.
Governance
Operates entirely within your domain, providing you with full control and oversight over data security and access.
Simplicity
Optimal model produced in one training run with a single API call.
EXPLAINABLE NEURAL NETS
XNNs are transparent hyper-graph neural models organised and orchestrated symbolically, integrating the power of neural nets with the observability of symbolic logic.
At inference (query), an XNN communicates its prediction along with a precise explanation of how that prediction was formed in a single forward pass.
XNNs are encapsulated in Explanation Structure Models that provide the symbolic fabric that makes the models fully interpretable.
Transparent
The behaviour of each XNN component can be observed, measured, and anticipated with absolute certainty.
Auditable
The query explanation combined with the precise model map produces a complete and immutable audit record for each prediction.
Robust
Activation mapping of XNN components facilitate game-changing drift and anomaly detection, measurement, and mitigation.
DECISION WORKFLOWS
Real-time explanations create the programmatic and process space that actionably separates the model prediction from the implemented decision.
This decision space allows for the programmatic codification of strategies, rules, requirements, and preferences to assure fit-for-purpose decisions.
Typical decision workflows include rules to identify, assess, and resolve any sub-optimal characteristics of a prediction prior to consuming the decision.
Identify
Test the prediction using explanation data against some codification of intent (e.g. identify bias).
Assess
Establish the severity of an identified concern in a prediction to inform the choice of resolution.
Resolve
Apply a ‘fix’ to align a prediction to established standards and include a human-in-the-loop when needed.
XNN PLATFORM
A simple developer library with full documentation, examples, and support for interactive and notebook systems.
Deploys on customer cloud (native cloud support), so no client data transfer is required.
Supported by an ecosystem of technology and solution development partners.
Efficient and Easy to use
The data onboarding process includes the automatic generation of pre-processing pipelines minimizing the effort and skills needed to engineer your data.
Reliable and Robust
Hybrid Intelligence solutions are completely predictable, so you can qualify, test and audit systems with the same reassurance as standard software, making them suitable for the most demanding applications.
Lifecycle Management
The transparent nature of XNNs and the surgical nature of the induction process supports an efficient, effective, and sensible new paradigm of model lifecycle management.