|
Random projection is a non adaptive dimension reduction technique where data matrix (tensor of order 2) is linearly embedded into much lower dimension by multiplying with a random matrix. This transformation is also an isometry with very high probability.
To implement this classical technique for tensor of higher order; this random matrix consumes huge memory even for tensors of moderate size along each mode. In this work we develop a memory efficient random projection technique along with theoretical and experimental guarantees.
|
