State-of-the-Art Innovations
to Prevent Financial Risk

The Feedzai Research department invests in applied research to improve our products and help users have a better experience. We work closely with Product and Customer Success to develop and transfer innovations. We focus on long-term, disruptive, state-of-the-art research, produce and protect our IP, publish peer reviewed work, contribute to open-source, partner with external researchers, and sponsor scholarships.

Recent Publications

AutoVizuA11y: a Tool to Automate Screen Reader Accessibility in Charts

Diogo Duarte, Rita Costa, Pedro Bizarro, and Carlos Duarte

Published at EuroVis 2024

PDF | GitHub | npm | YouTube

DiConStruct: Causal Concept-based Explanations through Black-Box Distillation

Ricardo Moreira, Jacopo Bono, Mário Cardoso, Pedro Saleiro, Mário A. T. Figueiredo, Pedro Bizarro

Published at CLeaR 2024 - Conference on Causal Learning and Reasoning

arXiv

On the Importance of Application-Grounded Experimental Design for Evaluating Explainable ML Methods

Kasun Amarasinghe, Kit T. Rodolfa, Sérgio Jesus, Valerie Chen, Vladimir Balayan, Pedro Saleiro, Pedro Bizarro, Ameet Talwalkar, Rayid Ghani

Published at AAAI-24 - Annual AAAI Conference on Artificial Intelligence

arXiv

FiFAR: a fraud detection dataset for learning to defer

Jean V. Alves, Diogo Leitão, Sérgio Jesus, Marco O. P. Sampaio, Pedro Saleiro, Mário A. T. Figueiredo, Pedro Bizarro

Published at ICAIF23 - Synthetic Data for AI in Finance workshop

arXiv | GitHub

AutoVizuA11y: a Tool to Automate Screen Reader Accessibility in Charts

Diogo Duarte, Rita Costa, Pedro Bizarro, and Carlos Duarte

Published at EuroVis 2024

PDF | GitHub | npm | YouTube

DiConStruct: Causal Concept-based Explanations through Black-Box Distillation

Ricardo Moreira, Jacopo Bono, Mário Cardoso, Pedro Saleiro, Mário A. T. Figueiredo, Pedro Bizarro

Published at CLeaR 2024 - Conference on Causal Learning and Reasoning

arXiv

On the Importance of Application-Grounded Experimental Design for Evaluating Explainable ML Methods

Kasun Amarasinghe, Kit T. Rodolfa, Sérgio Jesus, Valerie Chen, Vladimir Balayan, Pedro Saleiro, Pedro Bizarro, Ameet Talwalkar, Rayid Ghani

Published at AAAI-24 - Annual AAAI Conference on Artificial Intelligence

arXiv

FiFAR: a fraud detection dataset for learning to defer

Jean V. Alves, Diogo Leitão, Sérgio Jesus, Marco O. P. Sampaio, Pedro Saleiro, Mário A. T. Figueiredo, Pedro Bizarro

Published at ICAIF23 - Synthetic Data for AI in Finance workshop

arXiv | GitHub

Recent Blog Posts

Building Trust in a Digital World: The Role of Machine Learning in Behavioral Biometrics

In the world of financial services, the bank or financial institution’s relationship with the customer relies on digital trust, which is anchored in two fundamental principles. First, it must ensure the person engaging through digital banking channels is genuinely the individual they claim to be. Second, it must confirm that this person is authorized to complete the intended financial transaction.

Javier Liébana

AML Reimagined: LaundroGraph Exploits Graph Structure to Assist Anti-Money Laundering Activities

We introduce LaundroGraph, a self-supervised system based on graph-neural networks to assist experts during anti money laundering

Mário Cardoso

TimeSHAP: Explaining recurrent models through sequence perturbations

Recurrent Neural Networks (RNNs) are a family of models used for sequential tasks, such as predicting financial fraud based on customer behavior. These models are very powerful, but their decision processes are opaque and unintelligible to humans and rendering them black boxes to humans. Understanding how RNNs work is imperative to assess whether the model is relying on any spurious correlations or discriminating against certain groups. In this blog post, we provide an overview of TimeSHAP, a novel model-agnostic recurrent explainer developed at Feedzai. TimeSHAP extends the KernelSHAP explainer to recurrent models. You can try TimeSHAP at Feedzai’s Github.

Joao Bento, André Cruz, Pedro Saleiro

Building Trust in a Digital World: The Role of Machine Learning in Behavioral Biometrics

In the world of financial services, the bank or financial institution’s relationship with the customer relies on digital trust, which is anchored in two fundamental principles. First, it must ensure the person engaging through digital banking channels is genuinely the individual they claim to be. Second, it must confirm that this person is authorized to complete the intended financial transaction.

Javier Liébana

AML Reimagined: LaundroGraph Exploits Graph Structure to Assist Anti-Money Laundering Activities

We introduce LaundroGraph, a self-supervised system based on graph-neural networks to assist experts during anti money laundering

Mário Cardoso

TimeSHAP: Explaining recurrent models through sequence perturbations

Recurrent Neural Networks (RNNs) are a family of models used for sequential tasks, such as predicting financial fraud based on customer behavior. These models are very powerful, but their decision processes are opaque and unintelligible to humans and rendering them black boxes to humans. Understanding how RNNs work is imperative to assess whether the model is relying on any spurious correlations or discriminating against certain groups. In this blog post, we provide an overview of TimeSHAP, a novel model-agnostic recurrent explainer developed at Feedzai. TimeSHAP extends the KernelSHAP explainer to recurrent models. You can try TimeSHAP at Feedzai’s Github.

Joao Bento, André Cruz, Pedro Saleiro

Page printed in 18 Jul 2024. Plase see https://research.feedzai.com for the latest version.