AI & Functional Safety: Out-of-Distribution Detection as a Safety Mechanism

Date
2027-05-31
Location
Online
Host
Zach L

About this event

A live 30-minute expert session on Out-of-Distribution Detection as a Safety Mechanism (AI / ML — AI & Machine Learning Safety). What We'll Cover: What OOD detection is and why distributional shift is the dominant failure mode for deployed ML Core methods step by step: softmax confidence, energy scores, autoencoder reconstruction, and ensembles — and how to pick How an OOD monitor maps to a safety mechanism claim in ISO 26262/ISO 21448 terms and the detection-rate evidence it produces Common mistakes: tuning thresholds on in-distribution data only and ignoring detector latency in the fault-reaction time budget The false-negative/false-positive characterization and ODD-shift test evidence an auditor expects Related topics: out of distribution detection · ood detection · distributional shift

Topics

Registration

View event page

About Functional Safety Foundations

This online session delivers expert-led instruction on the regulatory landscape and implementation approach for Functional Safety Foundations.

The cross-sector fundamentals underlying every functional safety standard: risk as severity × likelihood, the safety function concept, integrity levels, the safety lifecycle, and the split between random and systematic failures.

What it regulates

Shared vocabulary and reasoning: hazard → risk → risk reduction allocation → safety requirement → verified implementation → maintained integrity in operation.

Who must comply

Everyone entering the field — the conceptual base of the entire catalog.

Key requirements of Functional Safety Foundations

  • Hazard and risk fundamentals
  • Safety function anatomy: sensor-logic-actuator with defined safe state
  • Integrity ladders (SIL/PL/ASIL/DAL) as risk-reduction currencies
  • Random vs systematic failure treatment
  • Lifecycle and management of functional safety incl. competence

Key concepts: Functional Safety Foundations

Functional safety management
Planning, competence, documentation and assessment wrapped around technical work.
Random vs systematic
Hardware wear-out/chance failures quantified statistically versus design/process errors controlled by rigor.
Risk reduction allocation
Distributing required risk reduction among mechanical design, safeguards, control functions and procedures.
Safe state
The defined condition a function drives the equipment to on demand or fault.

Frequently asked questions: Functional Safety Foundations

Why can't testing alone prove safety?

Rare-event targets (like 1e-7/h) are unverifiable by test duration, and systematic errors evade random testing — hence the standards' dual machinery of quantified hardware analysis plus process rigor and independent assessment.

Standard information based on the published text of Functional Safety Foundations. Event content is provided by the host. For authoritative guidance, consult the official standard body.