A team of data experts, passionate about the challenges and technologies of the Time Series
The Lab's missions
Monitoring of new methods or algorithms to stay on the cutting edge of technology in forecasting, classification, clustering, and anomaly detection
Knowledge management to accelerate the development of future projects thanks to our reusable source code base
Detailed and documented packages on topics such as data quality, preprocessing, and modeling
Our research topics
Forecasting
By developing benchmarks we compare the performance of different use cases such as power consumption forecasting, wind productivity and call center activity forecasting. This is made possible by testing and researching the performance with classical static methods (LASSO, GAM...), machine learning methods (random forest, boosting...), or expert aggregates. But also with more complex approaches, notably via Deep Learning (LSTM, transformers...).
Classification/Clustering
With a repertoire supporting the development of future projects (customer segmentation, clustering of load curves or dimension reduction). But also through writing best practice guides and articles on the main steps of a clustering project.
Anomaly detection
The variety of methods being important, feedback on data quality missions or anomaly detection via sensors (industrial, IoT, etc.) allows us to capitalize on our knowledge. And therefore to direct research on the methods likely to be the most efficient (ML, probabilistic...).
Boosting search engine capabilities of RegReview:…
RegReview is an AI solution for compliance teams, to automate regulatory monitoring and processes, which brings together several tools, the most essential of which is a search engine operating on a compiled database of custom-built regulatory sources.
We propose to explore several keyword extraction techniques to label text clusters obtained after a Text Clustering or a Topic Modeling pipeline. This work is following our previous articles about Topic Modeling and Text Clustering (here and here).
For a period of six months, 5 students from Centrale Supélec and ESSEC worked collaboratively with Sia Partners on building a Python library to create fake - which we'll call synthetic - data.
But what's the point of creating fake data? How could it help organizations?