The « Finance and Insurance Reloaded » (FaIR) Interdisciplinary Research Program (IRP) of the Institut Louis Bachelier aims to foster innovations driven by new technologies (from Artificial Intelligence to Blockchains) in the financial and insurance sectors. One component of this program is a series of workshops hosted by the ACPR.

Following a series of exploratory works shedding light on the issues of explainability and governance of AI, the ACPR recently issued a discussion paper « Governance of Artificial Intelligence in Finance » focused on anti-money laundering and combating the financing of terrorism, internal risk models, and customer protection.

The insurance and financial industries are currently exploring numerous ways to ensure confidence and trust in the promising results achieved by machine learning (and more generally AI): customization of products and personalization of customer experience, risk intermediation (via increased automation and parsimonious models in high dimension), usage of alternative data (from satellite images to text), etc. In doing so, they face challenges such as understanding and controlling the biases produced by algorithms, or ensuring their robustness.

The most advanced regulators like the ACPR are taking part in the debate, discussing how to efficiently evaluate ML models over the course of their lifecycle, how to adapt the governance of business processes involving AI, and how to audit those processes.

Topics covered by this ILB-FaIR workshop will include the explainability of ML models, their smooth and orderly functioning in production, and the early identification of potential quality issues in the datasets (including model re-learning or on-the-fly learning). Benchmark datasets and baseline models will also be discussed as methods for the internal or external audit of ML. The discussion will sidestep into topics such as algorithmic biases, as well as adversarial methods (such as GANs) which are a candidate for improving the robustness of ML. 

Gathering international experts on these domains around the table will feed the public consultation opened by the ACPR and running until September 4.

The workshop will be introduced by Olivier Fliche (Head of Fintech-Innovation, ACPR).

Two key speakers will present their recent research:

  • Yannick Léo (Emerton Data) on the biases of machine learning
  • Xin Guo (UC Berkeley) on the stakes of adversarial learning.

    The roundtable will be moderated by Charles-Albert Lehalle (Capital Fund Management, Paris and Imperial College, London) and our speakers will be joined by:

  • Blanka Horvath (Kings’ College and Turing Institute, London)
  • Michel Crouhy (Head of Research & Development, Natixis)
  • Laurent Dupont (Data Scientist, ACPR).
Réservez votre billet

Lieu

En ligne