Yanlei Diao awarded an ERC Proof of Concept grant

Digital data has become ubiquitous in many activities. The number and complexity of this “big data” has exploded in recent years. One of the challenges is to be able to analyze data streams in real time to extract relevant information. Thanks to previous European funding (an ERC Consolidator grant for the BigFastData project), Yanlei Diao and his team have developed algorithms dedicated to such analyses, capable of detecting anomalies in data series, and providing explanations for them. 

“The ExplainableAD project aims to go a step further by integrating these algorithms into a fully functional research prototype, evaluating them in real-life use cases in order to understand, and exploring possible technology transfers” says the researcher.

One example is financial fraud, particularly for banks. Among the numerous data circulating on transactions, the system first identifies those that appear anomalous. Then, the system can provide an explanation: the time, volume, origin or destination may be unusual. Analysts can then decide what action to take. “In the global landscape of fully operational anomaly detection systems, our system is the only one that offers these explanations,” says Yanlei Diao.

The main challenge lies in the fact that applications abound but have different properties. As a result, “normal” behaviors are very diverse. There is also the difficulty of separating anomalies from noisy data. While financial and banking applications are the first to be tackled by the ExeplainableAD project, others have already been identified, such as IT systems management, or healthcare.

About ERC Proof of Concept grants

ERC Proof of Concept grants are designed to extend a research project already funded by a European Research Council grant. These grants are aimed at facilitating the exploration of projects' potential for commercial and social innovation, by supporting complementary studies.

*LIX: a joint research unit CNRS, École Polytechnique, Institut Polytechnique de Paris, 91120 Palaiseau, France