The n-vector transformation is described in section 2.3. Gregor et al. 2017 use support vector regression and random forests, not neural networks. They compare to previous existing neural net that does not use coordinates as a feature.
If I remember correctly, I think somebody mentioned using coordinates as a feature isn't ideal? I think It was Mike or Pierre who brought this up?
All,
Here is the paper I mentioned that uses an n-vector to represent coordinate space: Gregor et al. 2017 Biogeosciences
The n-vector transformation is described in section 2.3. Gregor et al. 2017 use support vector regression and random forests, not neural networks. They compare to previous existing neural net that does not use coordinates as a feature.
If I remember correctly, I think somebody mentioned using coordinates as a feature isn't ideal? I think It was Mike or Pierre who brought this up?
-Luke