RESEARCH AND DEVELOPMENT 

[ TEAMS ]

​Internal AI Lab

​University chair with CentraleSupelec

CentralSupelec student teams

Polytechnique student teams

LEARN MORE

[ UNIVERSITY CHAIR ]

​4 Years

​2 PhDs Thesis

3 Teacher - Researcher

30 Students

AI Lab Team

Major Partner Bank

Powerful Computing Resources

​    - Jean Zay Supercomputer (CNRS)
    - CentraleSupelec Mesocentre (CNRS)

​Site of the university Chair : https://chaire-lusis.centralesupelec.fr/en/

LEARN MORE

[ FRAUD RESEARCH CHALLENGES ]

ML Fraud detection faces:

Data imbalance with less than 0.5% frauds

Concept drift with payments behaviors and fraud changing over time

Explainability of model prediction

LEARN MORE

[ TRADING RESEARCH CHALLENGES ]

Machine Learning Market prediction faces:

​Stochastic processes

Aleatoric and epistemic uncertainty

Post prediction events influence

Sensitivity to initial conditions

Standard ML metrics are insufficient

LEARN MORE

“It’s one of the most difficult problems in applied machine learning.”

Ciamac Moallemi, professor at Columbia Business School

---

“ Using machines to beat the markets is a really difficult challenge,
But I don’t think it’s impossible. ”

Jon McAuliffe, professor at the University of California at Berkeley and  CIO at Voleon Capital Management LP

[ RESEARCH 2021 - 2022 ]

- Rules extraction from credit card fraud detection ML model by using Minimum Description Length approach
- Advanced supervised Deep Learning models  built for tabular imbalanced data (ResNet, FTT, XBNet) compared to GBDT
- TracInAD  : measuring influence for Anomaly Detection (Thesis), article published to IJCNN 2022 (Rank A conference) 
- Anomaly detection  : 7 models (OCSVM, DeepSVDD, Autoencoder, Goad, RaPP, IF, NeutralAD) compared to GBDT (Thesis),  article under review for Springer special issue on imbalanced learning

- Robustness analysis of deep reinforcement learning state of the art methods for portfolio allocation (Thesis), article published in OkIP CAIS 2022 conference
- Deep reinforcement learning for stock portfolio allocation
- Measure uncertainty of a regressor model trained on market time series
- Probabilistic ML for trading
- Automated features search with Kernel Density Estimator

- Collaborative Deep Learning improvement using features extracted from sound files in addition to metadata

LEARN MORE

[ RESEARCH 2020 - 2021 ]

- Rules extraction from credit card fraud detection ML model
- Memory networks applied to credit card fraud for addressing concept drift
- Resampling 18 state of the art methods compared to raw GBDT. Article published to arXiv
- Categorical encoding impact of 7 state of the art encoding compared. Article published to arXiv

- Probabilistic ML quantifying aleatoric and epistemic uncertainty for market prediction
- Hawkes processes-based ML for market predictions
- Time Series patterns clustering from a latent space representation

- Quantum ML exploration and application

LEARN MORE

[ RESEARCH 2019 - 2020 ]

- Credit card fraud detection with deep learning
- Apply graph-based machine learning  to credit card fraud detection
- Explainability  of Deep Learning based models for credit card fraud detection

- Predict price direction with Convolutional Networks
- Learn trading strategy with deep reinforcement learning
- Create a metric reflecting the real expected performances of a market direction predictor
- Optimize trade exit inside the prediction horizon of existing AI based trading strategies running in production
- Optimize combination of existing AI based trading strategies running in production

- Music similarity from sound spectrograms
- Neural collaborative filtering for recommender

LEARN MORE

Research and Development
LUSIS AI

CONTACT US