Granted Patent B2

US 12,567,957 · App. 18/645,954 · Granted Mar 3, 2026

Inventor: Gilad Katz (Tel Aviv, IL)

Assignee: B.G. NEGEV TECHNOLOGIES AND APPLICATIONS LTD., AT BEN-GURION UNIVERSITY

H04L9/0861G09C5/00H04L9/0662

Abstract

Claims (87)

1 . An organization's system configured to label a given document based on an on-cloud classification service, while maintaining confidentiality of the given document's content from all entities external to the organization, comprising:

a. an encoder configured to receive said given document, and to create an embedding of the given document;

b. a deconvolution unit having a neural network, wherein weights of neurons within the neural network are defined relative to a key, said deconvolution unit being configured to receive said embedding, deconvolve the embedding, thereby to create a scrambled document which is then sent to the on-cloud classification service; and

c. a pre-trained internal inference network, configured to: (i) receive from said on-cloud classification service a cloud-classification of said scrambled document; (ii) also receive a copy of said embedding; and (iii) identify, given said received cloud-classification and said embedding copy, said label of said given document.

2 . The system of claim 1 , wherein said embedding is a reduced size of said given document, and wherein said scrambled document is of increased size compared to said embedding.

3 . The system of claim 1 , wherein a type of said given document is selected from text, table, and image.

4 . The system of claim 1 , wherein the internal inference network is a machine-learning network that is trained by: (i) a plurality of documents and respective true labels, and (ii) a plurality of respective cloud classifications resulting from submission each of the plurality of said documents, respectively, to a portion of the system that includes said encoder, said deconvolution unit, and said cloud classification service.

5 . The system of claim 1 , wherein said key is periodically altered, and wherein said internal inference network is re-trained upon each key alteration.

6 . The system of claim 1 , particularly adapted for labeling a text document, wherein:

said text document is separated into a plurality of sentences;

each sentence is inserted separately into said encoder as a given document; and

said pre-trained internal inference network identifies a true label of each said sentences, respectively.

7 . The system of claim 1 , particularly adapted for labeling a given table-type document, wherein:

said encoder has the form of a row/tuple to image converter;

said encoder receives at its input separately each row of said given table-type document; and

said pre-trained internal inference network identifies a true label of each said rows, respectively.

8 . The system of claim 1 , wherein:

additional documents, whose labels are known, respectively, are fed into said encoder, in addition to said given document;

a concatenation unit is used to concatenate distinct embeddings created by the encoder for said given document and said additional documents, thereby forming a combined vector V;

said combined vector V is fed into said deconvolution unit; and

said pre-trained internal inference network is further configured to: (i) receive, in addition to the copy of said embedding, a label of each said additional documents;

and (ii) identify said true label of said given document based on the labels of each said additional documents, in addition to said received cloud-classification, and said embedding copy.

9 . A method enabling an organization to label a given document based on an on-cloud classification service, while maintaining confidentiality of the given document's content from all entities external to the organization, comprising:

a. encoding said given document, resulting in an embedding of the given document;

b. deconvolving said embedding by use of a deconvolution unit comprising a neural network, wherein weights of neurons within the neural network are defined relative to a key, thereby to create a scrambled document, and sending the scrambled document to the on-cloud classification service;

c. using a pre-trained internal inference network to: (a) receive from said on-cloud service a cloud-classification of said scrambled document, (b) to also receive a copy of said embedding, and (c) to identify, given said received cloud-classification and said embedding copy, a true label of said given document.

10 . The method of claim 9 , wherein said embedding is a reduced size of said document, and wherein said scrambled document is of increased size compared to said embedding.

11 . The method of claim 9 , wherein a type of said given document is selected from text, table, and image.

12 . The method of claim 9 , wherein the internal inference network is a machine-learning network that is trained by (i) a plurality of documents and respective true labels, and (ii) a plurality of cloud classifications resulting from said encoding, deconvolution, and transfer of same documents, respectively, through said cloud classification service.

13 . The method of claim 9 , further comprising periodically altering said key, and further re-training said internal inference network upon each key alteration.

14 . A multi-organization system for commonly training a common on-cloud classification service by labeled given documents submitted from all organizations, while maintaining confidentiality of the documents' contents of each organization from all entities external to that organization, comprising:

a training sub-system in each organization comprising:

a. an encoder configured to receive a given document, and to create an embedding of the given document;

b. a deconvolution unit having a neural network, wherein weights of neurons within the neural network are defined relative to a key, said deconvolution unit being configured to receive said embedding, deconvolve the embedding, thereby to create a scrambled document which is then sent for training to the common on-cloud classification service, together with the respective label of that given document.

15 . The multi-organization system of claim 14 , wherein upon completion of the common training by labeled documents from all organizations, said common on-cloud classification service is ready to provide confidential documents classification to each said organizations.

16 . The multi-organization system of claim 14 , wherein during real-time labeling of new documents, each organization's sub-system comprising:

a. an encoder configured to receive a new un-labeled document, and to create an embedding of the new document;

b. a deconvolution unit having a neural network, wherein weights of neurons within the neural network are defined relative to a key, said deconvolution unit being configured to receive said embedding, deconvolve the embedding, thereby to create a scrambled document which is then sent to the on-cloud classification service;

c. a pre-trained internal inference network, configured to: (a) receive from said on-cloud service a common cloud-classification vector of said scrambled document, (b) to also receive a copy of said embedding, and (c) to identify, given said received common cloud-classification vector and said embedding copy, a true label of said un-labeled document.

17 . The system of claim 14 , wherein said embedding is a reduced size of said new document, and wherein said scrambled image is of increased size compared to said embedding.

18 . The system of claim 14 , wherein a type of said document is selected from text, table, and image.

19 . The system of claim 16 , wherein the internal inference network of each organization is a machine-learning network that is trained by a plurality of documents and respective true labels, and a plurality of respective common cloud classification vectors resulting from said encoding, deconvolution, and submission to the common cloud classification service.

20 . The system of claim 14 , wherein said key in each organization is periodically altered, and each organization's internal inference network is re-trained upon each key alteration.

21 . An organization's system configured to label a given document based on an on-cloud classification service, while maintaining confidentiality of the given document's content from all entities external to the organization, comprising:

a. a first encoder configured to receive said given document, and to create an embedding of the given document;

b. a deconvolution unit having a neural network, wherein weights of neurons within the neural network are defined relative to a key, said deconvolution unit being configured to receive said embedding, deconvolve the embedding, thereby to create a scrambled document which is then sent to the on-cloud classification service;

c. a pre-trained internal inference network, configured to: (a) receive from said on-cloud service a cloud-classification of said scrambled document, (b) to also receive a copy of said embedding, (c) to also receive activations vector reflecting activations created at the deconvolution unit during transfer of the embedding through it, and (d) to identify, given said received cloud-classification, said embedding copy, and said activations vector, a true label of said given document;

wherein said key is a unique key which is randomly generated for each document.

22 . The system of claim 21 , wherein said activations vector is a vector compressed relative to the entire activations created during the passage of the embedding through the deconvolution unit, and wherein said compression is performed by a second encoder.

23 . The system of claim 22 , wherein said second encoder is a trained or untrained encoder.

24 . The system of claim 21 , wherein said embedding is a reduced size of said given document, and wherein said scrambled document is of increased size compared to said embedding.

25 . The system of claim 21 , wherein a type of said given document is selected from text, table, and image.

26 . The system of claim 21 , wherein the internal inference network is a machine-learning network that is trained by: (i) a plurality of documents embeddings and respective true labels, (ii) said activations vectors, respectively, and (iii) a plurality of respective cloud classifications resulting from submission each of the plurality of said documents, respectively, to a portion of the system that includes said first encoder, said deconvolution unit, and said cloud classification service.

27 . The system of claim 21 , particularly adapted for labeling a text document, wherein:

said text document is separated into a plurality of sentences;

each sentence is inserted separately into said first encoder as a given document; and

said pre-trained internal inference network identifies a true label of each said sentences, respectively.

28 . The system of claim 21 , particularly adapted for labeling a given table-type document, wherein:

said first encoder has the form of a row/tuple to image converter;

said first encoder receives at its input separately each row of said given table-type document; and

said pre-trained internal inference network identifies a true label of each said rows, respectively.

29 . The system of claim 21 , wherein:

additional documents, whose labels are known, respectively, are fed into said first encoder, in addition to said given document;

a concatenation unit is used to concatenate distinct embeddings created by the first encoder for said given document and said additional documents, thereby forming a combined vector V;

said combined vector V is fed into said deconvolution unit; and

said pre-trained internal inference network is configured to: (a) receive from said on-cloud service a cloud-classification of said scrambled document, (b) to also receive a copy of said embedding, and a label of each said additional documents; and (c) to identify a true label of said given document based on said received cloud-classification, the labels of each said additional documents, and said embedding copy.

30 . A method enabling an organization to label a given document based on an on-cloud classification service, while maintaining confidentiality of the given document's content from all entities external to the organization, comprising:

a. encoding said given document, resulting in an embedding of the given document;

b. deconvolving said embedding by use of a deconvolution unit comprising a neural network, wherein weights of neurons within the neural network are defined relative to a key, thereby to create a scrambled document, and sending the scrambled document to the on-cloud classification service; and

c. using a pre-trained internal inference network to: (a) receive from said on-cloud service a cloud-classification of said scrambled document, (b) to also receive a copy of said embedding, (c) to also receive activations vector reflecting activations created at the deconvolution unit during transfer of the embedding through it, and (d) to identify, given said received cloud-classification, said embedding copy, and said activations vector, a true label of said given document;

wherein said key is a unique key which is randomly generated for each document.

31 . The method of claim 30 , wherein said activations vector is a vector compressed relative to the entire activations created during the passage of the embedding through the deconvolution unit, and wherein said compression is performed by a second encoder.

32 . The method of claim 30 , wherein said embedding is a reduced size of said document, and wherein said scrambled document is of increased size compared to said embedding.

33 . The method of claim 30 , wherein a type of said given document is selected from text, table, and image.

34 . The method of claim 30 , wherein the internal inference network is a machine-learning network that is trained by (i) a plurality of documents and respective true labels, and (ii) a plurality of cloud classifications resulting from said encoding, deconvolution, and transfer of same documents, respectively, through said cloud classification service.

35 . A multi-organization system for commonly training a common on-cloud classification service by labeled given documents submitted from all organizations, while maintaining confidentiality of the documents' contents of each organization from all entities external to that organization, comprising:

a training sub-system in each organization comprising:

a. a first encoder configured to receive a given document, and to create an embedding of the given document;

b. a deconvolution unit having a neural network, wherein weights of neurons within the neural network are defined relative to a key, said deconvolution unit being configured to receive said embedding, deconvolve the embedding, thereby to create an activations vector which is then sent for training to the common on-cloud classification service, together with the respective label of that given document;

wherein said key is a unique key which is randomly generated for each document.

36 . The multi-organization system of claim 35 , wherein upon completion of the common training by labeled classification vectors from all organizations, said common on-cloud classification service is ready to provide confidential documents' classifications to each said organizations.

37 . The multi-organization system of claim 36 , wherein during run-time labeling of new documents, each organization's sub-system comprising:

a. a first encoder configured to receive a new un-labeled document, and to create an embedding of the new document;

b. a deconvolution unit having a neural network, wherein weights of neurons within the neural network are defined relative to said key, said deconvolution unit being configured to receive said embedding, deconvolve the embedding, thereby to create an activations vector which is then sent to the on-cloud classification service, which given the activations vector, returns the label of the document.

38 . The system of claim 35 , wherein said on-cloud classification service, during training, further receives scrambled documents created by the deconvolution unit, and during run-time, the on-cloud classification service also further receives scrambled documents that are created by the deconvolution unit.

39 . The system of claim 35 , wherein said embedding is a reduced size of said new document, and wherein said scrambled image is of increased size compared to said embedding.

40 . The system of claim 35 , wherein a type of said document is selected from text, table, and image.

Assignments (1)

Priority Claims (1)

Continuity (2)

References Cited (19)