I want to compare accuracy among these four methods in terms of reconstruction. These four methods are mentioned below:
1. EMD (empirical mode decomposition)
2. DMD (dynamic mode decomposition)
3. QN (quasi newton method)
4. GAN (generative adversarial network)
It would be highly appreciated if anyone provides detailed explanation along with some publications.
I would like someone to suport my answer, but I would say 3 QN (quasi newton method) and 2 DMD (dynamic mode decomposition)
Hi Mª Angeles Zorrilla Lopez-Perea,
Thanks for your answer. Could you please provide some more details in favour of your answer?
1 is a very vague option. It barely does not require any options, but this is why it is not chosen at al. 2 covers a large scope of options and it may be used for almost anything. The trouble is not ways is the best option to cover almost anything. 3 is better but not the large scope of 2. 2 is the best and 3 if you can't use 2. 4 is barely used. Far too specific and may make mistakes. It is not used at al. Very recently it could be used for almost anything, but it has far too many mistakes. We still use 2 and 3. Although we want the power of 4. 2 and 3 for common usage and barely no restrictions
The first one is not as accurate. The next two are very accurate. ELM?is halfway between EMD and the second option. In accuracy and speed(?). Normally speed is the reverse of accuracy but not the fourth option. The third option still is very accurate and does keep the accuracy and precision and speed. The fourth is almost impossible. You can still use but very limited.
Thanks a lot. Do you mean to say that both classifican & reconstruction accuracy are better than sparse ELM autoencoder?
Thanks a lot. Do you mean to say that both classifican & reconstruction accuracy of DMD are better than sparse ELM autoencoder.
When we say better we mean as a whole. Unless all the mistakes at the fourth. It is nonsense to sacrifice almost anything for just the little help at the last mode. But yes it is true we normally mean accuracy (and of course mistakes, if you want to name it precision). Truth is we normally pay attention to the mistakes than to anything else. Tell me if we are wrong at the very least. Accuracy yes we gain a lot but it has not much improvement for 1 (for 1.5 just a little, but this much we can have the vast majority providing I am beyond the scope of "common")
Thanks for clarification. Now, I am concerned about non-Euclidean system. Is DMD enough to handle non Euclidean system if we tend towards reconstruction accuracy? Real system is not like plane or surface which is Euclidean in nature. Do we need to incorporate geometric deep learning to retrofit DMD if we want to reconstruct accurately from a corrupted system? Or we need to blend both DMD & geometric deep learning to fully exploit high resolution spatio- temporal data from a non Euclidean system? Please provide clear explanation .
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts