It's always amazing to hear Professor Amari speak at an event Next opportunity:

In his 40s, after inventing the Hopfield network 10 years ahead of Hopfield, legend Professor Amari got bored with neural networks and decided to revolutionize information geometry

Had a lot of fun giving a talk at TokyoAI tonight, thanks for having me!

In reinforcement learning, why do we always maximize the discounted sum of rewards? Sum is a recursive function, but so are min, max, variance, and many others. That means we can estimate and optimize them directly with value-based and actor-critic methods.