CAMELOT

autonomiC plAtform for MachinE Learning using anOnymized data

The CAMELOT project expects to revolutionize the detection of financial fraud through Machine Learning techniques. The project aims to develop an innovative machine learning platform, which will tackle three key issues that hinder the efficiency and accuracy of modern AI applications such as machine learning models, cloud resources, anonymized data, privacy issues, and integrating information from different, independent, and heterogeneous data platforms.

Keywords: Fraud detection · Machine Learning · Anonymization

Leading company in Portugal
FEEDZAI, S.A.

Project Start Date
07/01/2020

Project End Date
07/01/2022

Using AI to improve the detection of financial fraud

Feedzai uses AI to detect and prevent financial fraud such as card fraud, money laundering, or account opening fraud for all types of financial institutions and large merchants. To support its clients, Feedzai manages about 2000 machines and spends about 500,000 eur/month in cloud computing resources.

The goals of the CAMELOT project are to improve the efficiency, accuracy, or quality of the machine learning platform and associated stack using efficient cloud resources.

What to expect

Feedzai expects to be able to sustain its aggressive company growth goals. Among the desired potential impacts of the project are to reduce by 50% the cost for tuning the platform’s configuration; 50% reduction in of the operational costs at steady state and 50% faster deployment times for new models when reusing models.

Meet the partners

Promoter:
FEEDZAI – Pedro Bizarro

Academic Co-promoters:
Universidade de Coimbra – Bruno Cabral
Faculdade de Ciências da Universidade de Lisboa – Alcides Fonseca
Instituto Superior Técnico – Paolo Romano

CMU:
Computer Science Department – David Garlan

Goals

The project will tackle three key issues that hinder the efficiency and accuracy of modern AI applications:

Ensuring real-time constraints during both the training and inference phases of machine learning models while minimizing operational costs deriving from cloud resources.
Enabling learning over anonymized data, thus circumventing the privacy issues that currently prevent the reuse of information across models trained on datasets belonging to various entities (e.g., different financial institutions).
Integrating information from different, independent, and heterogeneous data platforms (e.g., key-value stores, relational and graph databases) in an automatic approach that maximizes machine learning applications’ performance.