> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.umnai.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.umnai.com/_mcp/server.

# Concepts

Umnai models are designed to make predictions and expose the reasoning behind those predictions.

The concepts section explains the model structures, explanation types, and analysis workflows that make this possible. It connects the core ideas behind Umnai’s modelling approach with the views and outputs available through the platform.

## How the concepts fit together

Umnai models are created through induction, a training process that discovers the structure of an Explainable Neural Net, or XNN.

An XNN makes predictions and exposes the reasoning behind them through modules, partitions, rules, attributions, activations, and decompositions. An Explanation Structure Model, or ESM, organises this information into structured explanations.

Those explanations are exposed through views, which provide the API outputs used for inspection, auditing, visualisation, and downstream workflows.

## Explore by topic

| Topic                 | Concepts                                                                                                                                                                                                                                                                             |
| --------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| Model structure       | [Explainable Neural Net (XNN)](./explainable-neural-net-xnn), [Explanation Structure Model (ESM)](./explanation-structure-model-esm), [Induction](./induction)                                                                                                                       |
| Modelling approach    | [Neuro-symbolic AI and hypergraph models](./neuro-symbolic-ai-and-hypergraph-models), [Symbolic reasoning](./symbolic-reasoning)                                                                                                                                                     |
| Explanation types     | [Global explanations](./explanations/global-explanations/overview), [Local explanations](./explanations/local-explanations/overview), [Context explanations](./explanations/context-explanations)                                                                                    |
| Explanation workflows | [Model behaviour](./using-explanation-components/model-behaviour), [Identifying bias](./using-explanation-components/identifying-bias), [Model monitoring](./using-explanation-components/model-monitoring), [Model optimisation](./using-explanation-components/model-optimisation) |

## Model concepts

Model concepts describe the structures behind Umnai models.

[**Explainable Neural Net (XNN)**](./explainable-neural-net-xnn) explains the predictive model structure, including modules, partitions, rules, bins, decompositions, and the model intercept.

[**Explanation Structure Model (ESM)**](./explanation-structure-model-esm) explains how XNN outputs are organised into structured explanations.

[**Induction**](./induction) explains how Umnai creates model structure from training data.

[**Neuro-symbolic AI and hypergraph models**](./neuro-symbolic-ai-and-hypergraph-models) explains the broader modelling approach: combining learned patterns with symbolic structures that can be inspected and reasoned over.

[**Symbolic reasoning**](./symbolic-reasoning) explains reasoning strategies behind symbolic systems, including deduction, induction, and abduction.

## Explanation concepts

Explanation concepts describe the types of explanations the platform can produce.

[**Global explanations**](./explanations/global-explanations/overview) describe model behaviour as a whole: the patterns found during induction, the model’s structure, and the components that influence predictions across a dataset.

[**Local explanations**](./explanations/local-explanations/overview) describe a specific prediction or batch of predictions, including attributions, activations, decision actions, and strength or weakness signals.

[**Context explanations**](./explanations/context-explanations) add comparison or scenario context to local explanations, such as what-if analysis, control swaps, or nearest-neighbour comparisons.

## Explanation workflows

Explanation workflows show how explanation components support practical analysis.

They cover model behaviour, bias analysis, attribution justification, data segment identification, low data coverage, monitoring, optimisation, real-time explanations, and audit logs.

## Decision Space

[**Decision Space**](./decision-space) describes the operating space in which model decisions and explanations can be understood.

It provides a higher-level view of how data, model behaviour, and explanations relate to the decisions supported by the platform.