AI Technical Risk Controls

AI systems are increasingly used in contexts where reliability, safety, fairness, privacy and trust matter. In this course, you’ll learn how to assess the robustness of neural networks using the structured approach set out in ISO/IEC TR 24029-1:2021. You’ll explore why non-robust systems can behave unexpectedly, create unfair outcomes or become vulnerable to adversarial attacks, and you’ll learn how to identify and evaluate these risks before deployment. Through clear explanations, worked examples and practical assessment scenarios, you’ll examine statistical, formal and empirical methods for testing robustness. You’ll learn how to set robustness goals, choose suitable datasets and metrics, define thresholds, interpret results and document evidence-based decisions. The course also introduces key techniques such as data perturbation and abstract interpretation, helping you understand how neural networks respond to changed, distorted or deliberately manipulated inputs.

Learn how to evaluate machine learning classification models with confidence using the principles and guidance of ISO/IEC TS 4213:2022. This course takes you beyond model development to focus on what matters most: determining whether a classifier is accurate, reliable, efficient, and fit for purpose. Drawing on real-world AI applications, you will explore the complete classification assessment process, from understanding data quality, bias, and evaluation design to selecting and interpreting the most appropriate performance metrics. You will learn how to assess binary, multi-class, and multi-label classification models using measures such as accuracy, precision, recall, F-scores, ROC and Precision-Recall curves, AUROC, AUPRC, Hamming Loss, Jaccard Index, and Kullback-Leibler divergence. The course also explores operational considerations including latency, throughput, computational efficiency, and energy consumption, as well as statistical techniques for comparing model performance and validating results. What makes this course unique is its standards-based approach, combining technical evaluation methods, operational performance measures, and statistical significance testing within a single practical framework. By the end of the course, you will be able to evaluate classifiers rigorously, compare competing models objectively, and communicate performance results with greater confidence and credibility.