BoundLab Documentation#
BoundLab is a framework for building neural network verification tools using symbolic expressions and bound propagation.
Mental model: represent a set of inputs as an Expr (a sum of typed components over shared error symbols), transform it with an abstract interpreter (for example zono.interpret), then concretize into sound bounds (lb, ub, lbub).
This documentation is organized into three learning paths:
Getting Started: install BoundLab and run your first bound computation.User Guide: understand the expression system, propagation APIs, and interpreter workflow.Examples: end-to-end patterns you can adapt to your own models.
Install BoundLab and run your first verification script.
Core concepts, supported operations, and practical workflow guidance.
Task-focused examples from manual expression building to model interpretation.
Autogenerated API docs for all public modules, classes, and functions.