Today, we’re thrilled to announce the next innovation in our commitment to delivering Responsible AI: Feedzai’s FairGBM. It is a game-changing algorithm that makes Fair Machine Learning accessible to all. FairGBM is available in our products and through an open-source release for non-commercial uses.
How FairGBM Enables Responsible AI
Right this minute, there’s a bank using a biased algorithm that will deny someone a much-needed loan. The decision is not because the applicant isn’t qualified. They are. It’s because they have the “wrong” zip code to be approved for a loan. At Feedzai, this doesn’t sit well with us. So we did what we were born to do – we developed innovations to solve a problem.
FairGBM represents a new significant Responsible AI breakthrough. Feedzai’s FairGBM is a general purpose algorithm that trains models to simultaneously optimize both predictive performance and fairness. In other words, banks don’t have to choose between fair models or models that perform well. FairGBM can be trained with millions of data points while optimizing for both goals. Fairness and accuracy now go hand in hand.
But we haven’t even gotten to the best part: we’re open sourcing it for non-commercial uses!!
Why We’re Open Sourcing Feedzai’s FairGBM
FairGBM was initially developed for banks, FinTechs, and other financial institutions. But FairGBM is generic by design. This means that while it was developed by Feedzai, it isn’t limited to financial services and financial crime prevention. It can be used by any organization that is committed to delivering model fairness at scale regardless of decision-making task or who their end-users are.
When we realized that FairGBM can be used by any domain with any type of data, that’s when the spark was ignited. We knew we could advance Responsible AI practice by making this important breakthrough accessible and available to not only Feedzai customers through our products, but to the rest of the world through an open-source release for non-commercial uses. FairGBM has the potential to improve the AI practice of numerous organizations outside of financial services, including how healthcare providers, government agencies, and non-profit groups deliver their services.
FairGBM open-source release is available in our FairGBM github repo. We invite you to try it!
How FairGBM Works
We’ve invested time, resources, and passion into mitigating AI bias and I’m happy to say, it’s working. Last spring, we introduced Feedzai Fairband, an AutoML algorithm that automatically discovers less biased machine learning models with zero additional model training costs. Fairband went on to win a slew of awards. It also introduced Responsible AI to many leaders in the financial services sector and proved how it’s both accessible and attainable for financial services organizations worldwide.
Fairband tweaks the hyperparameter optimization process to find the fairest model possible but does not enforce models to be fair by design. FairGBM is a constrained version of gradient boosted trees that optimizes for both performance and fairness between groups. And it does so without sacrificing predictive performance.
Because it was built on the LightGBM framework, FairGBM’s is fast and scalable to train models using millions of data points, an essential requirement for financial services.
Feedzai’s Commitment to Customer Success and Fairness
At Feedzai, we are 100% committed to our customer’s success. It’s the “why” for all the innovations we develop. Our customers are under pressure from their customers to consider environmental, social, and governance (ESG) issues in their business practices. Providing FairGBM in our products is just one way we ensure our clients’ success. Our clients can demonstrably prove to their customers that they have Responsible AI frameworks, and are treated fairly based on their actions – not their zip code, age, gender, or race.
We demonstrate our commitment to Responsible AI with the launch of Fairband. Developing FairGBM and making it open source for non-commercial uses is the next logical step in our efforts to make financial services fairer and to advance Responsible AI for everyone.
FairGBM ushers in a new era of AI fairness. We can’t wait to see how it makes financial services – and many other industries – fairer for everyone!
If you want to learn more about Feedzai’s world-changing research and innovations visit Feedzai Research.
Share this article:
Pedro Bizarro
Pedro Bizarro is Co-founder and Chief Science Officer at Feedzai, where he manages the Feedzai Research department. Pedro work focuses on building the conditions, teams, culture, and quality for the long-term impact of applied research at Feedzai in the areas of AI, Data Visualization, and Systems. Pedro is an avid runner and Ironman.
Related Posts
0 Comments15 Minutes
ChatGPT Revolutionizes Retail Banking…Soon
The world is buzzing with excitement about Large Language Models (LLM) and Generative AI…
0 Comments11 Minutes
ChatGPT Creates New Data Science Challenges
ChatGPT can spit out a fraudulent email from Amazon in a matter of seconds. It can pass…
0 Comments8 Minutes
Understanding FairGBM: Feedzai’s Experts Discuss the Breakthrough
Feedzai recently announced that we are making our groundbreaking FairGBM algorithm…