Ajay Thampi - Interpretable AI: Building explainable machine learning systems

Ive been fortunate to have worked with data and machine learning for about a decade now. My background is in machine learning, and my PhD was focused on applying machine learning in wireless networks. I have published papers ( http://mng.bz/zQR6 ) at leading conferences and journals on the topic of reinforcement learning, convex optimization, and classical machine learning techniques applied to 5G cellular networks.

After completing my PhD, I began working in the industry as a data scientist and machine learning engineer and gained experience deploying complex AI solutions for customers across multiple industries, such as manufacturing, retail, and finance. It was during this time that I realized the importance of interpretable AI and started researching it heavily. I also started to implement and deploy interpretability techniques in real-world scenarios for data scientists, business stakeholders, and experts to get a deeper understanding of machine-learned models.

I wrote a blog post ( http://mng.bz/0wnE ) on interpretable AI and coming up with a principled approach to building robust, explainable AI systems. The post got a surprisingly large response from data scientists, researchers, and practitioners from a wide range of industries. I also presented on this subject at various AI and machine learning conferences. By putting my content in the public domain and speaking at leading conferences, I learned the following:

• I wasnt the only one interested in this subject.

• I was able to get a better understanding of what specific topics are of interest to the community.

These learnings led to the book that you are reading now. You can find a few resources available to help you stay abreast of interpretable AI, like survey papers, blog posts, and one book, but no single resource or book covers all the important interpretability techniques that would be valuable for AI practitioners. There is also no practical guide on how to implement these cutting-edge techniques. This book aims to fill that gap by first providing a structure to this active area of research and covering a broad range of interpretability techniques. Throughout this book, we will look at concrete real-world examples and see how to build sophisticated models and interpret them using state-of-the-art techniques.

I strongly believe that as complex machine learning models are being deployed in the real world, understanding them is extremely important. The lack of a deep understanding can result in models propagating bias, and weve seen examples of this in criminal justice, politics, retail, facial recognition, and language understanding. All of this has a detrimental effect on trust, and, from my experience, this is one of the main reasons why companies are resisting the deployment of AI. Im excited that you also realize the importance of this deep understanding, and I hope you learn a lot from this book.

Writing a book is harder than I thought, and it requires a lot of workreally! None of this would have been possible without the support and understanding of my parents, Krishnan and Lakshmi Thampi; my wife, Shruti Menon; and my brother, Arun Thampi. My parents put me on the path of lifelong learning and have always given me the strength to chase my dreams. Im also eternally grateful to my wife for supporting me throughout the difficult journey of writing this book, patiently listening to my ideas, reviewing my rough drafts, and believing that I could finish this. My brother deserves my wholehearted thanks as well for always having my back!

Next, Id like to acknowledge the team at Manning: Brian Sawyer, who read my blog post and suggested that there might a book there; my editors, Matthew Spaur, Lesley Trites, and Kostas Passadis, for working with me, providing high-quality feedback, and for being patient when things got rough; and Marjan Bace, for green-lighting this whole project. Thanks as well to all the other folks at Manning who worked with me on the production and promotion of the book: Deirdre Hiam, my production editor; Pamela Hunt, my copyeditor; and Melody Dolab, my page proofer.

Id also like to thank the reviewers who took the time to read my manuscript at various stages during its development and who provided invaluable feedback: Al Rahimi, Alain Couniot, Alejandro Bellogin Kouki, Ariel Gamio, Craig E. Pfeifer, Djordje Vukelic, Domingo Salazar, Dr. Kanishka Tyagi, Izhar Haq, James J. Byleckie, Jonathan Wood, Kai Gellien, Kim Falk Jorgensen, Marc Paradis, Oliver Korten, Pablo Roccatagliata, Patrick Goetz, Patrick Regan, Raymond Cheung, Richard Vaughan, Sergio Govoni, Shashank Polasa Venkata, Sriram Macharla, Stefano Ongarello, Teresa Fontanella De Santis, Tiklu Ganguly, Vidhya Vinay, Vijayant Singh, Vishwesh Ravi Shrimali, and Vittal Damaraju.Special thanks to James Byleckie and Vishwesh Ravi Shrimali, technical proofreaders, for carefully reviewing the code one last time shortly before the book went into production.