I have a matrix where each column represents a feature and each row represent a data point. I am trying to convert this matrix into a 1D vector which can contain the joint distribution among the features within the data. Does i-vectors work in this case and how does rolling out the upper triangular portion of covariance matrix work? Any insights or directions will help.
Hessam Fathipour Thank you for the idea. However, that was my first thought, to get the covariance matrix and unroll the upper triangular portion, similar to the approach taken for the final layer in basic CNNs. But I was wondering if there is any advanced method like SVD, to get that vector instead of depending on covariances.
Yes, you can do it via PCA/SVD or LDA etc., see e.g., https://stats.stackexchange.com/questions/192722/dimension-reduction-using-pca-in-matlab
Muhammad Ali Thank you, PCA works fine for 1 dimensional data where there is no question of spatial information. But, for me the 2D learning is important as the row and column locations are important.
In a nutshell, I am looking for an algorithm that is close to CNN but is not that data hungry
- Badges
- Science topic
- Similar topics
- Analytical Chemistry
- Column