Advances in Sparse Classification

A recent result in compressed sensing (CS) allows us to perform non-parametric speech recognition that is robust to noise, and that requires few training examples. By taking fixed length representations of training samples and stacking them in a matrix, we form a frame, or an over-complete basis. Gemmeke and Cranen have shown that sparse projections onto this frame recover the correct transcription with 91% accuracy at -5 dB SNR. We propose that the goal of speech recognition is not sparse projection onto training tokens, but onto training types. Sparse projection onto types can be achieved by building a frame for each word in the dictionary, and stacking the frames to form a rank 3 tensor. Speech recognition is performed by convex linear projection onto the tensor, with sparsity enforced only in the index that specifies type. We derive a mixed L1/L2 relaxation that can be globally optimized using Newton descent.unpublishednot peer reviewedU of I OnlyUndergraduate senior thesis not recommended for open acces