Reasoning on Natural Inputs

Explore a thought-provoking lecture on the challenges and opportunities of applying classical algorithms to real-world problems. Delve into the tension between abstract algorithm design and practical application, using the maximum flow problem as a case study. Discover how neural algorithmic reasoning offers a promising approach to bridge this gap, enabling reasoning on natural inputs. Learn about the potential of combining deep neural networks as feature extractors with classical algorithmic techniques to enhance problem-solving in reinforcement learning environments like Atari games. Gain insights from Petar Veličković of DeepMind Technologies as he presents cutting-edge research in this field, drawing from the historical context of algorithm design to modern advancements in machine learning.

Introduction
Why study algorithms
Flow networks
The problem with abstract algorithms
Algorithmic reasoning