MATHEMATICS

Abstract:Artificial intelligence, propelled by advancements in machine learning, has made significant strides in solving complex tasks. However, the current neural network-based paradigm, while effective, is heavily constrained by inherent limitations—chief among them being a lack of structural organization and a learning progression characterized by undesirable properties like catastrophic forgetting. As AI research advances without a unifying framework, it either attempts to heuristically patch weaknesses or loosely incorporates biological mechanisms without strong theoretical foundations. Meanwhile, a recent paradigm shift in evolutionary theory, driven primarily by advances in evolutionary developmental biology (EDB), has been largely overlooked in AI literature, despite a striking analogy between the previously dominant understanding of evolution, the Modern Synthesis, and contemporary machine learning. Consequently, the principles of adaptation derived from EDB, which have reshaped our understanding of the evolutionary process, can also provide the foundation for a unifying conceptual framework for the next design philosophy in AI, grounded firmly in biology’s first principles. 
In this talk, I will outline the analogy between the Modern Synthesis and modern machine learning, highlighting how core principles from EDB can inform a new design paradigm for AI. I will then present a novel learning approach built upon these principles—one that overcomes several major limitations of contemporary neural-network-based methods. Notably, this approach guarantees adaptation to a good solution without overparameterization, inherently preserves existing knowledge when learning new information, and provides a structured representation that is both comprehensible upon inspection and readily integrable with symbolic methods.

Bio: Zeki Doruk earned his undergraduate degree in Electrical and Electronics Engineering from Middle East Technical University (METU) and completed his Master’s in Robotics & Computational Neuroscience at École Polytechnique Fédérale de Lausanne (EPFL). He is currently a doctoral student at EPFL, focusing on a developmental design paradigm for AI systems.