Guided character string alteration
AssigneeInternational Business Machines Corporation
InventorsPablo Loyola, Kugamoorthy Gajananan, Yuji Watanabe, Fumiko Akiyama
Life of this patent
Estimating the term…
Abstract
Guided character string alteration can be performed by obtaining an original character string and a plurality of altered character strings, traversing the original character string with a first Long Short Term Memory (LSTM) network to generate, for each character of the original character string, a hidden state of a partial original character string up to that character, and applying, during the traversing, an alteration learning process to each hidden state of a partial original character string to produce an alteration function for relating partial original character strings to partial altered character strings.
Claims14
A computer program product including one or more non-transitory computer readable storage mediums collectively storing program instructions that are executable by a computer to cause the computer to perform operations comprising:
- obtaining an original character string and a plurality of altered character strings;
- traversing the original character string with a first Long Short Term Memory (LSTM) network to generate, for each character of the original character string, a hidden state of a partial original character string up to that character;
- and
- applying, during the traversing, an alteration learning process to each hidden state of a partial original character string to produce an alteration function for relating partial original character strings to partial altered character strings, thereby developing counter-defenses against attacks,
- wherein the traversing of the original character string further generating, for each character in the original character string, an output state of that character, and the applying of the alteration learning process further selecting an action based on each hidden state, and selecting an alternate character based on each output state corresponding to a hidden state,
- wherein the action is one of adding the alternate character to the original character, replacing the original character with the alternate character, permuting the original character with the previous character in the original character string, removing the original character, and maintaining the original character.
A computer-implemented method comprising:
- obtaining an original character string and a plurality of altered character strings;
- traversing the original character string with a first Long Short Term Memory (LSTM) network to generate, for at each character of the original character string, a hidden state of a partial original character string up to that character;
- and
- applying, during the traversing, an alteration learning process to each hidden state to produce an alteration function for relating partial original character strings to partial altered character strings, thereby developing counter-defenses against attacks,
- wherein the traversing of the original character string further generating, for each character in the original character string, an output state of that character, and the applying of the alteration learning process further selecting an action based on each hidden state, and selecting an alternate character based on each output state corresponding to a hidden state,
- wherein the action is one of adding the alternate character to the original character, replacing the original character with the alternate character, permuting the original character with the previous character in the original character string, removing the original character, and maintaining the original character.
An apparatus comprising:
- a memory device for storing program code;
- and
- a processor device operatively coupled to the memory device for running the program code to:
- obtain an original character string and a plurality of altered character strings;
- traverse the original character string with a first Long Short Term Memory (LSTM) network to generate, for at each character of the original character string, a hidden state of a partial original character string up to that character;
- and
- apply, during traversing, an alteration learning process to each hidden state to produce an alteration function for relating partial original character strings to partial altered character strings, thereby developing counter-defenses against attacks,
- wherein the traversing of the original character string further generating, for each character in the original character string, an output state of that character, and the applying of the alteration learning process further selecting an action based on each hidden state, and selecting an alternate character based on each output state corresponding to a hidden state,
- wherein the action is one of adding the alternate character to the original character, replacing the original character with the alternate character, permuting the original character with the previous character in the original character string, removing the original character, and maintaining the original character.
Dependencies are inferred from the claim text.