His research interests focus on AI-based automation techniques applied to root cause analysis and resource management in large dynamic networks. He received his Ph.D in signal processing and telecommunications at INRIA, University of Rennes, France.

Growing complexity in the hyper-connected society is a major challenge for system management and control. The digital system of the future will provide global connectivity to billions of IoT devices and end-users, and will rely on cloud-enabled telecommunication networks such as 5G and beyond. Traditional approaches to resource management based on system model specification will be increasingly defied due to the emergence of these complex and dynamically evolving networks.

On the other hand, instead of logically reasoning on the causes of observed data, today’s AI approaches teach the machine to associate patterns to specific meaning. This approach fails when the possibilities to learn are too many, and there is not enough data evidence to generalize over the unobserved possibilities. True machine intelligence must be built based on the ability to reason on why it is observing a specific phenomenon and the ability to derive from where it is originating. It can then focus the attention to causally relevant aspects of the ever-growing information flow, instantaneously understand the current situation, and take the most appropriate decisions.

In Bell labs, we focus on devising casual models that automatically provide explanations and diagnose complex networks. For controllable environments, we elaborate interventional learning techniques allowing the discovery of stimulus-reaction behaviors of networked components through injection of small perturbations of computing resources. For passively observed environments, we can still infer the causal relationships under certain assumptions in both operational and failure modes.

In addition, we explore a new management paradigm of continuous learning in interaction which brings a strong promise for highly adaptive control. We elaborate model-free Reinforcement Learning based algorithms allowing to hide the inherent complexity of the environment and adapt to its changing conditions. To overcome the convergence speed and safety challenges, these model-free approaches need to integrate with causal inference disciplines.