Propostas Submetidas

DEI - FCTUC
Gerado a 2024-11-21 20:44:53 (Europe/Lisbon).
Voltar

Titulo Estágio

Explainability of ML-based model in mission critical scenarios

Áreas de especialidade

Engenharia de Software

Sistemas Inteligentes

Local do Estágio

Coimbra

Enquadramento

Explainability of ML-based model outputs is of high importance to financial institutions for compliance and regulatory purposes. The proposed challenge is to compute explanations for ML algorithms such as tree-based ones (e.g.: LGBM) or deep learning ones, within mission-critical services constraints (strict SLAs and SLOs) and for multiple scenarios (e.g.: batch, runtime or on-demand). Extensive benchmarks will be required to validate all proposed architectures and implemented solutions.

Objetivo

Explainability of ML-based model outputs is of high importance to financial institutions for compliance and regulatory purposes. The proposed challenge is to compute explanations for ML algorithms such as tree-based ones (e.g.: LGBM) or deep learning ones, within mission-critical services constraints (strict SLAs and SLOs) and for multiple scenarios (e.g.: batch, runtime or on-demand). Extensive benchmarks will be required to validate all proposed architectures and implemented solutions.

Plano de Trabalhos - Semestre 1

"0. Onboarding on Feedzai's Model Serving
1. Literature review on available explainability algorithms for both Tree-based and Deep-Learning-based algorithms
2. Assess their viability regarding required compute power and latency -- some might not be viable for low-latency runtime scenarios
3. Assess existing explainability frameworks regarding batching support and/or optimizations
3. Discovery: Impact of on-demand explanations in product -- will require conducting interviews, etc.

Expected results:
- Literature review document
- Plan approach for second semester based on those findings
"

Plano de Trabalhos - Semestre 2

"1. Develop runtime Proof-of-Concept explanation service under Feedzai's ML Infra ecosystem
2. (stretch goal) Develop batch Proof-of-Concept explanatin service under Feedzai's ML Infra ecosystem
3. Benchmark and load-test those two services -- determine their performance in terms of throughput and latency


Expected results:
- Two explainability Proof-of-Concept services (real-time and batch)
- Assessment of their performance in throughput and latency (critical to determine their viability)
- Documentation, final report and presentation
"

Condições

Remunerated

Orientador

Alberto Ferreira
alberto.ferreira@feedzai.com 📩