Lake et al (2017)
I like a lot the interaction between development and artificial intelligence. I also really liked model-based and model-free methods in reinforcement learning. However, this distinction is not that clear to me. Can you explain it further?
The paper talks about 2 developmental start-up software: intuitive physics and intuitive psychology. I’m having difficulty understanding how early inductive biases can be integrated into deep neural networks. I'm also confused about how Bayesian models address the complexity of human cognition and these early start-up software. There are priors of priors of priors but at the very first place what is happening?
At some point of the paper they state: “We are less committed to a particular story regarding the origins of the ingredients, including the relative roles of genetically programmed and experience-driven developmental mechanisms in building these components in early infancy.” However, doesn’t understanding the origin and interaction of these ingredients seem central to capturing the complexity of human intelligence?
Rescorla & Wagner
I'm surprised that I didn't think about the conditioning and reinforcement learning literature from a probabilistic perspective before.
I think the model they talked about is also related to Lake et al ( 2017) paper's model-based reinforcement learning and planning action sequences to maximize future rewards. Is it? And can we talk about this notion more? Regarding the idea of learning occurring only when certain expectations are violated, I understand its efficiency, but I’m still unclear about how this principle operates in more complex scenarios.