Extracting data from the hash memory using an error key

Abstract

This paper addresses the problem of retrieving data from hash memory using erroneous keys. Propose a character-level Convolutional Neural Network (CNN) model that corrects misspelled input words before hash table lookup. Approach generates synthetic errors (deletion, insertion, transposition) with up to two modifications from a base vocabulary of 10 words. The model encodes input strings as character sequences, applies embedding and ID convolution layers, and classifies them into the correct word classes. Experimental results show high accuracy in correcting common typing errors, demonstrating the feasibility of using neural networks as a preprocessing layer for fault-tolerant hash memory systems.