Titulo Estágio
Reject Inference study
Áreas de especialidade
Engenharia de Software
Engenharia de Software
Local do Estágio
Coimbra
Enquadramento
Fraud is an adversarial problem where criminals routiney try to counteract defenses. Our typical approach is to train models on positive labels from e.g. chargebacks only, ignoring auto-declines. We'd like to diagnose how this affects our model performances, and explore mitigation strategies (e.g. using measures of model confidence to include some auto-declines as positives, ...).
Objetivo
Fraud is an adversarial problem where criminals routiney try to counteract defenses. Our typical approach is to train models on positive labels from e.g. chargebacks only, ignoring auto-declines. We'd like to diagnose how this affects our model performances, and explore mitigation strategies (e.g. using measures of model confidence to include some auto-declines as positives, ...).
Plano de Trabalhos - Semestre 1
"1. Onboarding
2. Literature review: label noise / incomplete labels / model confidence / conformal prediction / etc.
3. Data exploration
4. Define the experimental design
5. Data preparation"
Plano de Trabalhos - Semestre 2
"6. Implementation of selected methods
7. Execution of experiments
8. Reporting"
Condições
Remunerated
Orientador
João Costa
joao.viriato@feedzai.com 📩