Skip to content

Latest commit

 

History

History
26 lines (21 loc) · 1.24 KB

Towards a Formal Model of Type 1 Diabetes for Artificial Intelligence.md

File metadata and controls

26 lines (21 loc) · 1.24 KB

Introduction

  • T1D patients have to compute multiple times a day their dose, which is error prone.
  • Need for systems to be really robust before they can be deployed.
  • Recommender system for bolus dose is proposed.

Formal specification

  • Using Event-B formal modelling language - provides verification guarantees using set theory as a language.

  • Event B is limited to natural and integer numeric types.

  • Providing invariants and limits in the specification allows for a simple model to ensure a proper bolus dose.

  • Variables:

    • maxBolus
    • targetRangeUpper - target blood glucose upper value
    • targetRangeLower - target blood glucose lower value
  • Invariants:

    • maxBolus is a natural number <= 500 IU
    • targetRangeUpper is a natural number and >= 55 and <= 150
    • targetRangeLower is a natural number and >= 30 and <= 80
    • targetRangeLower <= targetRangeUpper
  • These examples illustrate critical safety constraints for a T1D bolus calculator involving AI can be captured formally. Note there are difficult challenges with verifying AI systems that modify themselves.