Stand up for ethical AI! How to detect and mitigate AI bias using Kubeflow
summary
Shows a working Kubeflow pipeline that plugs IBM’s libraries into KFServing to detect and mitigate bias: AI Fairness 360 runs 70+ fairness metrics and exposes 10+ mitigation algorithms, Adversarial Robustness 360 measures robustness (example: clean accuracy 85% drops to 15% on adversarial samples) and Explainability 360 produces LIME-style saliency. The demo uses payload logging via CloudEvents to collect requests, calculates disparate impact (examples shown at 0.77 and 0.5), and reuses flagged examples to create retraining data.
AI-written summary · about
chapters (6)
- 0:00Start
- 1:26Center for Open Source Data and Al Technologies
- 6:41Adversarial threats to
- 9:23Adversarial Robustness 360
- 21:30Payload Logging to enable
- 27:56LFAI Trusted AI Committee
related talks
28mLife is not fair, but your ML pipeline can be!
23mRemoving Unfair Bias in the Machine Learning Pipeline
40mEthical AI: Addressing Bias & Algorithmic Fairness
22mAI in Healthcare: Bias, Privacy, and Trust in Kube-Native Systems
11mKeynote: Trusted AI: Bringing Trust Back into AI through Open Source
47mFight AI Bias With… Bias
12mTrust and transparency of AI for the enterprise
47mEthics & AI: Identifying & preventing bias in predictive models