3 ms·
Honestly it seems a bit like stirring the linear algebra pile for the sake of stirring, I don’t think fitting parameters to observation is a space that needs de
by SiempreViernes 8y ago
Honestly it seems a bit like stirring the linear algebra pile for the sake of stirring, I don’t think fitting parameters to observation is a space that needs deep neural nets.
They talk about their method allowing fast experimentation with how to train dnn, but what is really needed computationally is fast n-body sims to explore the theory space not a fast classifier once you have a fully converged simulation.
I just can’t see what role the trained network has in cosmology.
- grumpopotamus 8y agoRecently there was some very interesting work on solving ambiguous inverse problems with Invertible Neural Networks: https://hci.iwr.uni-heidelberg.de/vislearn/inverse-problems-invertible-neural-networks/ https://hci.iwr.uni-heidelberg.de/vislearn/inverse-problems-... >We can for example say that this specific observation y either corresponds to a young cluster with large expansion velocity, or to an older system that expands slowly.