Ant Financial AI Advances Presented at ICML 2019

On June 9, 2019, the International Conference on Machine Learning (ICML) 2019 opened in Long Beach, USA, where Ant Financial showcased several accepted papers and hosted a finance‑focused AI workshop.

The workshop gathered AI experts who discussed cutting‑edge topics such as financial‑intelligent applications, “small data”, and data privacy security, aiming to address technical challenges in AI‑plus‑finance innovation.

After the workshop, many scholars continued lively exchanges on site.

Ant Financial’s AI team contributed multiple papers, including an adversarial user model for reinforcement‑learning‑based recommendation systems, a nonlinear distributional gradient temporal‑difference learning method, and a particle‑flow Bayes’ Rule algorithm for high‑dimensional Bayesian inference.

Below we briefly introduce three selected papers:

Adversarial User Model for Reinforcement Learning‑Based Recommendation System – The paper proposes using a generative‑adversarial user model as a simulated environment for offline RL training, dramatically reducing the need for online interaction data while enabling online policy updates based on real user feedback.

Nonlinear Distributional Gradient Temporal‑Difference Learning – The authors integrate distributional reinforcement learning with gradient temporal‑difference learning, introducing a Distributional Mean‑Squared Bellman Error objective that provides theoretical convergence guarantees and opens new research directions.

Particle Flow Bayes’ Rule – To tackle high‑dimensional Bayesian inference and online updating without storing large histories, the paper presents a particle‑flow Bayes operator based on ordinary differential equations, demonstrating superior accuracy and efficiency in several high‑dimensional experiments.

For more details, click “Read Original” to visit Ant Financial’s official website.