arXiv subjects

Frédéric Oblé

Publications and source records attributed to Frédéric Oblé.

3 recordsLinked to original sources

arXiv2021

Transfer Learning for Credit Card Fraud Detection: A Journey from Research to Production

The dark face of digital commerce generalization is the increase of fraud attempts. To prevent any type of attacks, state-of-the-art fraud detection systems are now embedding Machine Learning (ML) modules. The conception of such modules is only communicated at the level of research and papers mostly focus on results for isolated benchmark datasets and metrics. But research is only a part of the journey, preceded by the right formulation of the business problem and collection of data, and followed by a practical integration. In this paper, we give a wider vision of the process, on a case study of transfer learning for fraud detection, from business to research, and back to business.

cs.LG↗

arXiv2020

Master your Metrics with Calibration

Machine learning models deployed in real-world applications are often evaluated with precision-based metrics such as F1-score or AUC-PR (Area Under the Curve of Precision Recall). Heavily dependent on the class prior, such metrics make it difficult to interpret the variation of a model's performance over different subpopulations/subperiods in a dataset. In this paper, we propose a way to calibrate the metrics so that they can be made invariant to the prior. We conduct a large number of experiments on balanced and imbalanced data to assess the behavior of calibrated metrics and show that they improve interpretability and provide a better control over what is really measured. We describe specific real-world use-cases where calibration is beneficial such as, for instance, model monitoring in production, reporting, or fairness evaluation.

cs.LG↗

arXiv2015

Transforming Javascript Event-Loop Into a Pipeline

The development of a real-time web application often starts with a feature-driven approach allowing to quickly react to users feedbacks. However, this approach poorly scales in performance. Yet, the user-base can increase by an order of magnitude in a matter of hours. This first approach is unable to deal with the highest connections spikes. It leads the development team to shift to a scalable approach often linked to new development paradigm such as dataflow programming. This shift of technology is disruptive and continuity-threatening. To avoid it, we propose to abstract the feature-driven development into a more scalable high-level language. Indeed, reasoning on this high-level language allows to dynamically cope with user-base size evolutions. We propose a compilation approach that transforms a Javascript, single-threaded real-time web application into a network of small independent parts communicating by message streams. We named these parts fluxions, by contraction between a flow (flux in french) and a function. The independence of these parts allows their execution to be parallel, and to organize an application on several processors to cope with its load, in a similar way network routers do with IP traffic. We test this approach by applying the compiler to a real web application. We transform this application to parallelize the execution of an independent part and present the result.

cs.PL↗