Imagine you have 100 blue boxes. Each time you roll the dice, their shape changes. But all 100 boxes always share the same shape. If I understand correctly, in this situation the shape is the natural latent. While color is just static background information.
Imagine you have 100 boxes. Each time you roll the dice, their color changes. But all 100 boxes always share the same color. In this situation, color is the natural latent.
Imagine you have 100 boxes. Each time you roll the dice, their color changes. If at least one box is blue, all boxes are blue. Otherwise their color is independent. Is “all boxes are blue or all boxes have independent colors” a natural latent (it’s something you learn about all boxes by examining a single box)?
Does the latter (3) type of natural latents have any special properties, is it some sort of “meta-level” natural latent (compared to 2)? I’m asking because I think this type of latents might be relevant to how human abstractions work. Here’s where I wrote about it in more detail.
A question about natural latents.
Imagine you have 100 blue boxes. Each time you roll the dice, their shape changes. But all 100 boxes always share the same shape. If I understand correctly, in this situation the shape is the natural latent. While color is just static background information.
Imagine you have 100 boxes. Each time you roll the dice, their color changes. But all 100 boxes always share the same color. In this situation, color is the natural latent.
Imagine you have 100 boxes. Each time you roll the dice, their color changes. If at least one box is blue, all boxes are blue. Otherwise their color is independent. Is “all boxes are blue or all boxes have independent colors” a natural latent (it’s something you learn about all boxes by examining a single box)?
Does the latter (3) type of natural latents have any special properties, is it some sort of “meta-level” natural latent (compared to 2)? I’m asking because I think this type of latents might be relevant to how human abstractions work. Here’s where I wrote about it in more detail.