Explore the intricacies of implementing trusted AI through machine learning payload logging on Kubernetes in this conference talk by Tommy Li and Andrew Butler from IBM. Delve into the challenges of production model serving, examining KNative and KF Serving configurations, supported frameworks, and storage subsystems. Investigate the inference service control plane and KFServing deployment views with practical examples. Address the critical question of prediction trustworthiness in production ML architectures, focusing on payload logging. Discover the Linux Foundation's approach to the Trusted AI lifecycle through open source initiatives. Examine the importance of AI explainability, bias detection in criminal justice systems, and adversarial robustness. Learn about AI Explainability 360 and AI Fairness 360 tools, and explore LFAI Trusted AI projects integrated with Kubeflow Serving. Conclude with a demonstration that ties together the concepts presented.
Infusing Trusted AI Using Machine Learning Payload Logging on Kubernetes
Overview
Syllabus
Intro
Production Model Serving? How hard could it be?
KNative
KF Serving: Default and Canary Configurations
Supported Frameworks, Components and Storage Subsystems
Inference Service Control Plane
KFServing Deployment View
KF Serving Examples
Model Serving is accomplished. Can the predictions be trusted?
Production ML Architecture
Payload Logging Architecture Examples
Linux Foundation Al & Data
Trusted Al Lifecycle through Open Source
Al needs to explain its decisions!
Bias in Al: Criminal Justice System
Adversarial Robustness
Al Explainability 360
Al Fairness 360
LFAI Trusted Al Projects with Kubeflow Serving
Demo Flow
Taught by
Linux Foundation
