Borja Gonzalez Leon, PhD
Artificial General Intelligence, Open-endedness, Agents, Language, Games.
Founder, Principal Scientist & AI Lead at Iconic.
borjagleon [at] gmail [dot] com
Discovering Creative Behaviors through DUPLEX: Diverse Universal Features for Policy Exploration.Borja G. León, Francesco Riccio, Kaushik Subramanian, Peter Wurman & Peter Stone.
NeurIPS 2024.Reinforcement learning algorithm that learns diverse near-optimal policies in the GranTurismo 7 and generalises in far out-of-distribution physic settings while maintaining visually-diverse solutions.
Quality-Diversity Actor-Critic: Learning High-Performing and Diverse Behaviors via Value and Successor Features Critics.Luca Grillotti, Maxence Faldor, Borja G. León, Antoine Cully.
ICML 2024Quality-diversity reinforcement learning algorithm that learns diverse high-performing behaviors in continuous control tasks, enabling better adaptation to unexpected situations while maintaining both quality and behavioral diversity.
In a Nutshell, the Human Asked for This: Latent Goals for Following Temporal Specifications.Borja G. León, Murray Shanahan, Francesco Belardinelli
ICLR 2022Deep learning architecture that generates goal representations for multi-task temporal logic instructions, enabling better generalization to unseen tasks in out-of-distribution environments.
To be finished
I am a Founder, Principal Scientist and AI Lead at Iconic, where I lead research initiatives that will transform the landscape of AAA games. I am also a member of the European Laboratory for Learning and Intelligent Systems (ELLIS) and have been recognized in the Nova 2024 list of Top 11 Spanish tech professionals and the 2021 CIDOB 35 under 35 Leaders of the Future List. Previously, I spearheaded research projects at Sony AI and Meta, contributing to significant improvements in AI racing systems for Gran Turismo 7 and Ads targeting systems, respectively. I received my Ph.D. in Artificial Intelligence from Imperial College London under the supervision of Francesco Belardinelli and Murray Shanahan, where I focused on reinforcement learning and compositional generalisation of grounded language learning. My research has been published in top-tier conferences including NeurIPS, ICLR, and ICML. I am the founder and lead organizer of Imperial College Autonomous Reasoning & Learning (ICARL), an association sponsored by Google DeepMind, InstaDeep, and Iconic. My current work focuses on generally-capable human-like AI, with special attention to improv and dramatic acting in narrative-driven games.