We are proud of who we are.
For that reason, we join the global vindication of LGBTI+ rights and, in particular, the visibility of our Pride also in the workplace.

We are also proud of the things we do and what we are achieving. During the last few weeks, a team from BBVA AI Factory has been working on the design of a methodology inspired by peer review within the academic community that allows us to validate the models we build.

We all need at some point of the Machine Learning development cycle a new pair of eyes to have a look at our model and tell us what we might have missed.

This methodology divides every Data Science project into five phases and has been introduced at BBVA AI Factory through three pilot projects that cover the wide-ranging use cases present in our company: from projects aiming at building improved embeddings of card transactions, to using graphs in financial pattern analysis scenarios.

In each of these phases, the reviewed team -the team that has worked on the development of the project- provide the necessary documentation to the reviewer team -that is appointed at the beginning of the peer review process-.

In this post we explain how we came up with the first version of the methodology and what stages were followed in its design. We are also sharing a document that may be useful for other Data Science teams that want to implement a similar methodology.

Hey, we are one of the world's best financial innovation labs!

Global Finance has included the BBVA Artificial Intelligence Factory in its ‘World’s Best Financial Innovation Labs’ list for 2021. BBVA AI Factory SL is recognized as a benchmark in the application of artificial intelligence and data analytics in the creation and improvement of banking services worldwide.

Providing BBVA teams with key knowledge to detect fraud or accompanying our clients in the management of ther financial health through personalised messages and recommendations. These are just some of the areas we have been working on and which were submitted to the above awards.

One of the noteworthy developments we have launched in recent weeks, and which we included as part of our bid, was the Supplies and Subscriptions project, which allows our BBVA app clients in Spain to view their recurring utility bills and card subscriptions categorised by brand. In addition, a client can simulate how much money they could save in a year if they were to unsubscribe from any of these services.

This use case is relevant as these types of services represent an important part of client's fixed costs. We will talk more about this project in the coming months. Stay tuned!

Further reading

+Responsible Artificial Intelligence by design (BBVA -in spanish-)
At BBVA, the DATA, Client Solutions and Responsible Business teams have reflected on how to tackle the challenges of ethical artificial intelligence, integrating responsibility in a natural way in the main phases of an AI project: Ideation, data collection, system development and launch, control and feedback. There are different AI systems, understood as the European Commission's draft regulation does: from the one based on a linear regression model, which would be a simple system, to the one using a neural network, which would be a complex system. Responsibility applied to AI systems implies, regardless of the type of model used, putting people at the centre and maintaining rigour and caution.

Esther de la Torre, (Responsible Business area), has worked on the development of a practical guide for embedding responsibility in the development of services and products based on AI (PDF).

+A guide with requirements for Artificial Intelligence audits (AEPD - Spanish Data Protection Agency)
The document provides guidance and objective criteria, from a data protection perspective, that should be incorporated into audits of processing operations that include AI-based components.

+A Visual Exploration of Gaussian Processes (Distill)
With this article, the authors present an introduction to Gaussian processes and make the mathematical intuition behind them more approachable.

+Towards Human-Centered Explainable AI: the journey so far (The gradient)
According to Upol Ehsan, researcher, executive consultant and ex-Google & IBM Research, there is a myth surrounding the idea of "building a better AI", (techno centric assumption), which he calls the "algorithm-centered Explainable AI" myth. In the article, he challenge this myth and offer an alternative version of XAI, one that is sociotechnically informed and human-centered.

Quote

of the

month

"The successful future of research and data science is where small, purposeful data plays a big role."

Anahit Tevosyan, Director of Research, FINCA International

We love to tell what we do outside!