Deep Learning for Reliable Storage
Abstract
With the exponential increase of cloud based storage systems, it has become critical to reliably
store data. Traditionally, methods for error correction have relied on duplication of data / introduction
of artificial redundancy. Here, we leverage the natural redundancy present in the data using
deep learning based techniques. Deep learning is a subset of machine learning algorithms that have
given excellent results on a variety of tasks.
We describe DNN (deep neural net based) models for learning decompression in texts compressed
by Huffman coding. Firstly, we work with noiseless texts following which we work with
noisy texts. Next, we outline a model for bit erasure correction. For this, we present a DNN based
model for bit erasure correction in uncompressed, ASCII encoded texts. Finally, we describe a
model that does bit erasure correction for Huffman code compressed texts. Such an end-to-end
system can be useful for cases when the codebook / encoding algorithm is not available and decoding / error correction needs to be done.
Citation
Khurjekar, Ishan Dhananjay (2018). Deep Learning for Reliable Storage. Master's thesis, Texas A & M University. Available electronically from https : / /hdl .handle .net /1969 .1 /173974.