memory will be very large

[zdx3578]

If the data format is slightly larger, the memory will be very large. Do you have any suggestions for optimization? Any suggestions are welcome, thank you！

[ryanchankh]

On a high level, I think one potential way to reduce the memory is the learn (or construct) each E and C_j matrix by batch. As long as there is an efficiently way to solve the self-expressive regression problem at each step, then the construction of those matrices should be scalable.