Following on the idea that the theories we will need to tackle the complexity of the brain have not been developed yet (e.g. https://mastodon.social/@NicoleCRust/109472784550141853)
What types of up and coming theoretical(ish) frameworks are you most excited about? Dynamical systems / RNNs? Topology? Network theory? Something else entirely?
@complexsystems @cogneurophys @PessoaBrain @SussilloDavid @carlosbrody @Neurograce @neuralreckoning @tyrell_turing @DrYohanJohn @cian @WiringtheBrain @tdverstynen @neuralengine (Anyone?)
@NicoleCRust @complexsystems @cogneurophys @PessoaBrain @SussilloDavid @carlosbrody @Neurograce @neuralreckoning @DrYohanJohn @cian @WiringtheBrain @tdverstynen @neuralengine
My guess is we will see greater theoretical advances when we get a better mathematical hold on the deep connections between RL, control, and inference (real traction, not the hand-wavy active inference version we currently have).
Best version I've seen of this was from Sergey Levine:
https://arxiv.org/abs/1805.00909
Reinforcement Learning and Control as Probabilistic Inference: Tutorial and Review
The framework of reinforcement learning or optimal control provides a
mathematical formalization of intelligent decision making that is powerful and
broadly applicable. While the general form of the reinforcement learning
problem enables effective reasoning about uncertainty, the connection between
reinforcement learning and inference in probabilistic models is not immediately
obvious. However, such a connection has considerable value when it comes to
algorithm design: formalizing a problem as probabilistic inference in principle
allows us to bring to bear a wide array of approximate inference tools, extend
the model in flexible and powerful ways, and reason about compositionality and
partial observability. In this article, we will discuss how a generalization of
the reinforcement learning or optimal control problem, which is sometimes
termed maximum entropy reinforcement learning, is equivalent to exact
probabilistic inference in the case of deterministic dynamics, and variational
inference in the case of stochastic dynamics. We will present a detailed
derivation of this framework, overview prior work that has drawn on this and
related ideas to propose new reinforcement learning and control algorithms, and
describe perspectives on future research.
@NicoleCRust @complexsystems @cogneurophys @PessoaBrain @SussilloDavid @carlosbrody @Neurograce @neuralreckoning @DrYohanJohn @cian @WiringtheBrain @tdverstynen @neuralengine
For me, this is the key missing piece of theory because it could help us explain why a system largely focused on homeostasis (animals' bodies and their organs) evolved into a system that can do RL and which learns an internal model of the world.
Vicarious trial and error
When rats come to a decision point, they sometimes pause and look back and forth as if deliberating over the choice; at other times, they proceed as if they have already made their decision. In the 1930s, this pause-and-look behaviour was termed ‘vicarious ...
Nicole Rust
Blake Richards
Luiz Pessoa
Yohan John 🤖🧠
Jon Awbrey