About this blog

Hello! I’m Ahmed Akakzia, Senior Applied AI Engineer at InstaDeep, where I work on deploying cutting-edge AI systems in real-world industrial settings.

I hold a PhD in Artificial Intelligence from ISIR Lab, Sorbonne Université, completed under the supervision of Olivier Sigaud and Mohamed Chetouani. My doctoral thesis — Teaching Predicate-based Autotelic Agents: Learning Goal Representations with a Social Partner for Intrinsically Motivated Agents — investigated how artificial agents can form and pursue their own goals in open-ended environments.

My research interests span the study and design of embodied artificial agents that learn in open-ended, non-stationary environments. These agents must efficiently explore and discover skills without any external reward signal, relying instead on intrinsic motivation to represent, generate and pursue self-generated goals.

Two key ideas drive my work. First, learning abstract and task-agnostic goal representations — grounded in spatial semantic relations — promotes the transfer and adaptation of acquired skills to novel situations. Second, open-ended skill development is enriched not only by autonomous sensorimotor experience but also by socio-cultural signals from human partners. Integrating these two dimensions gives rise to Teachable Autotelic Agents — systems that learn autonomously and benefit from social guidance.

This blog is a portfolio of my scientific posts and publications.

Thank you for reading!