Explainable Neural Networks: Revolutionizing AI

"As far as AI in banking is concerned, explainability and fairness will be must-haves in 2-3 years because bill C-27 and OSFI expectations are going to require them." Ahead of the RE•WORK - Toronto AI Summit, we asked Eric Lanoix, Vice President, Quantitative Risk at Coast Capital Savings his thoughts on the topic. Here's what he had to say:

What do you think is the most important advancement for AI in Finance?

"Definitely explainable neural networks. They are revolutionizing AI, but somehow most practitioners in banking don’t know about them. That needs to change because, in high-model risk applications like credit adjudication, the regulator (OSFI) is raising the bar on the ability of model developers to explain algorithmic decisions. Post-Hoc techniques like LIME and SHAP have limitations and can be fooled. But you can’t fool a fully explainable neural network. I will discuss this and provide examples during my presentation."

What are some recent wins from an AI project you are working on? What challenges did you face during it? How did you overcome them?

"Getting our institution to accept decisions made by an AI was not easy, but we succeeded and our first neural network-based credit scorecard is currently in the late stages of independent validation. More of them are in development. The main challenge we faced is getting upper management to accept the risk of using a neural network. I don’t want to say too much about how we managed to gain acceptance because this is perhaps the coolest part of my presentation (no spoiler please J). I can simply say that having a continuous and fully explainable neural network was key to achieving that."

How engrained will AI be in the Finance space 5 to 10 years from now? What will some future trends be then?

"AI in credit underwriting is here to stay and its usage will only grow over time. That’s why the regulator (OSFI) is paying a lot of attention to it by refreshing its guideline on model risk management (E-23). As far as AI in banking is concerned, explainability and fairness will be must-haves in 2-3 years because bill C-27 and OSFI expectations are going to require them. Many banks are not paying too much attention to this, but they need to start to."

What are you most looking forward to while speaking at the AI Toronto Summit?

"Really looking forward to meeting other banking AI experts, exchanging with them, and learning from them."

Want to learn more?

Eric will be speaking at the upcoming Toronto AI Summit on November 9-10, 2022, at The Westin Harbour Castle, in Toronto, ON. Join him and many other AI/ML and Deep Learning experts to learn more about the latest trends and opportunities in the industry.

Standard Passes end on Friday, October 28, so book your place today. Use Promo Code: SAVE20 for 20% off your pass.

For more information, contact hello@re-work.co.