Название: Model Performance Management with Explainable AI: Build High Performance Responsible AI
Автор: Amit Paka, Krishna Gade, Danny Farah
Издательство: O’Reilly Media, Inc.
Год: 2021-11-22
Страниц: 82
Язык: английский
Формат: epub
Размер: 10.1 MB

Artificial Intelligence (AI) has the potential to provide productive, efficient, and innovative solutions to everyday problems. But it comes with risks. Multiple examples of alleged bias in AI have been reported in recent years, and many people were already affected by the time those issues surfaced. This could have been avoided if humans had visibility into every stage of the system life cycle.

These issues might have been avoided if humans had visibility into every stage of the system life cycle. Part of that life cycle involves training a Machine Learning (ML) model to help in making decisions. In the model validation stage, teams could have unearthed instances of unwanted model behavior. With visibility into model performance online and offline, these sorts of unwanted behaviors can be detected and managed early on.

For each high-profile case that comes under public scrutiny, there are probably many systems that are silently operating and negatively impacting lives. One of the main concerns with AI today is that issues are detected after the fact, usually when people have already been affected by them. This is a foundational problem in AI that needs a foundational fix. That’s where model performance management (MPM) comes in.

Technology teams often utilize DevOps principles and application performance management (APM) to rapidly iterate software development and ensure high performance, respectively. MPM is a framework that draws on these learnings and applies them to the unique challenges of working with ML models to provide teams with control and visibility over the entire ML workflow. MPM is to MLOps what APM is to DevOps: it’s a framework that offers observability and control over model performance.

Explainable AI (XAI) is a form of AI that aims at creating Machine Learning models that are, for the most part, explainable and/or interpretable by humans. XAI evolved out of the need to break open the black box of AI models to make them interpretable by humans, with the intent of minimizing the risk of unknown or unpredictable outcomes from those models. XAI is not only relevant for regulatory and legal reasons, but it is also an important tool for monitoring and managing model performance.

Contents:
Preface
1. Introduction to Model Performance Management and Explainability
DevOps and Application Performance Management
The Rise of MLOps
Model Performance Management
2. Explainable AI
Explainability in Context
Who Needs XAI?
How XAI Can Be Used to Manage Model Performance
XAI in Different Domains
3. The Machine Learning Life Cycle
The Three Types of Analytics
Life Cycle Stages
How MPM Fits into the Life Cycle
4. MPM in the ML Life Cycle
5. Implementing MPM in Practice
6. MPM and Responsible AI
About the Authors