Technical Insights

What is Neuro-Symbolic AI?

Neuro-symbolic AI combines the ability of neural systems to learn statistical patterns with the explicit concepts, rules and relationships used in symbolic reasoning. The objective is not to place two independent systems side by side, but to create an architecture in which learnt patterns can be named, inspected and used within structured reasoning.

What is Hybrid Intelligence?

Hybrid Intelligence is UMNAI's neuro-symbolic framework for building decision models whose predictions can be examined through explicit modules, partitions, rules, attributions and symbolic relationships.

What are Explainable Neural Networks?

Explainable Neural Networks (XNNs) are modular predictive models that build explanation into their computation. Each module represents a feature or interaction, each partition defines a region of behaviour, and each active rule produces a contribution that can be traced to the final output.

What is an Explanation Structure Model?

An Explanation Structure Model (ESM) is the explainable package around an XNN. It connects the data transformations, predictive model, symbolic identities, relationships and metadata needed to execute a query and organise its explanation.

What Makes a Good AI Explanation?

A useful AI explanation must reflect what the model actually did, contain enough information for its purpose and remain traceable to the relevant inputs and model components. It must also be presented at the right level for the person using it.

The Different Types of Explanations in Hybrid Intelligence

"Explain this result" can mean several different things. Hybrid Intelligence distinguishes seven explanation types so that the system answers what the reader actually asked, rather than returning the same feature-importance chart for every situation.

How XNN Attributions Explain Decisions

An Explainable Neural Network (XNN) attribution is a signed contribution produced by an active module rule. Attributions build the prediction and the explanation together: they show how much each feature or interaction pushed the model output up or down.

How to Interpret XNN Results

An Explainable Neural Network (XNN) result contains more than a label or number. A disciplined reading starts with the task and output scale, reconciles the background value and attributions, inspects the active rules, and then considers activation context and verification metadata.

How XNNs Support What-If and Counterfactual Analysis

Explainable Neural Networks (XNNs) can rerun the same explicit model logic under controlled changes. What-if analysis tests a change chosen by the user; counterfactual analysis searches for a feasible route to a target; what-if-not analysis isolates the effect of removing or replacing a factor.

How to Query XNNs and Receive Explanations

Querying an Explainable Neural Network (XNN) is an explainable inference workflow. The Explanation Structure Model (ESM) applies the recorded preprocessing, the XNN produces a prediction and module attributions, and the explainer assembles the rules, activations, views and metadata needed for the use case.

How XNN Predictions are Built

Explainable Neural Network (XNN) predictions are additive before the final task-specific transformation. Each active module rule produces an attribution, those contributions are combined with a background value, and an identity, sigmoid or softmax function produces the model output.

How Feature Rollups, Rule Views and Attribution Views Work

Explainable Neural Network (XNN) views answer different questions about the same execution. An attribution view shows contribution, a rule view shows the logic that activated, and a feature rollup brings direct and interaction effects back to the original inputs.

How Local, Module-Level and Global Interpretability Work

Explainable Neural Network (XNN) interpretability operates at three connected scopes. Local views explain one prediction, module-level views examine one feature or interaction, and global views aggregate behaviour across many queries.

The Hybrid Intelligence Induction Process

Induction is the process that learns the structure of a Hybrid Intelligence model. It analyses the data and optional knowledge, identifies useful feature interactions, constructs modules and partitions, generates local rules and packages the trained Explainable Neural Network (XNN) within an Explanation Structure Model (ESM).

Training and Retraining XNNs

Training optimises the parameters of an induced Explainable Neural Network (XNN). Retraining reuses a compatible module, partition and rule structure with updated data or objectives. The lifecycle retains a trainable XNN and produces a serving XNN for runtime use.

Retraining vs Reinduction

Retraining updates parameters within a compatible induced Explainable Neural Network (XNN) structure. Reinduction learns a new structure. The decision depends on whether the model's modules, partitions, inputs and assumptions still describe the problem being solved.

Monitoring Drift, Activation Paths and Explanation Stability

Explainable Neural Network (XNN) monitoring can observe not only inputs and outputs, but also which modules, partitions and rules are used. Changes in activation paths and attributions provide an additional view of drift and explanation stability.

Incorporating Human Knowledge into XNNs

Human Knowledge Incorporation connects explicit expert or organisational knowledge with the learnt structure of an Explainable Neural Network. Rules, constraints, taxonomies and causal models can be represented symbolically, reviewed and managed alongside data-derived behaviour.

Auditing XNN Outputs

Auditing an Explainable Neural Network output means reconstructing the relevant decision record: what was submitted, which model ran, what it predicted, which components contributed, what happened afterwards and whether the retained information still matches its integrity reference.

Unique Verification Codes

Unique Verification Codes are cryptographic hashes calculated from selected information about an Explainable Neural Network model or query. They provide tamper-evident fingerprints that support later integrity checks.

Privacy-Preserving Auditability

Auditability does not require every reviewer to see every raw field. A privacy-preserving design separates the evidence needed for integrity and lineage from the sensitive data needed only for authorised investigation.

Trusted AI Infrastructure

Trusted AI infrastructure is not a single feature. It is the connected system that preserves lineage from data and knowledge through induction, deployment, runtime explanation, monitoring, version change and independent audit.