Reasoning Algorithmically- From Toy Experiments to AGI Modules

Explore the transformative potential of neural networks capable of executing algorithmic computation in this 52-minute lecture by Petar Veličković from DeepMind Technologies. Delve into the journey from isolated toy experiments to groundbreaking applications in detecting patterns in complex mathematical objects and supporting generalist reasoning agents. Gain insights into the intersection of graphs, classical algorithms, reasoning, and intelligent systems as Veličković shares his personal account of the field's rapid development. Discover how these advancements could revolutionize both machine learning and theoretical computer science, enabling extrapolative generalization and expanding the applicability of classical algorithms to previously inaccessible inputs.