Method for determining hidden states of systems using privacy-preserving distributed data analytics

• US 9,246,978 B2
• Filed: 11/11/2013
• Issued: 01/26/2016
• Est. Priority Date: 11/11/2013
• Status: Active Grant

• Current Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Inventors: Wang, Ye, Xie, Qian, Rane, Shantanu
• Primary Examiner(s): Sall, El Hadji
• Application Number:
  US14/076,307
• Publication Number:
  US 20150134778A1
• Time in Patent Office: 806 Days
• Field of Search: 709/219, 709/203, 709/200, 709/223, 707/736, 706/20, 706/45, 706/62, 704/256, 704/256.1, 704/273
• US Class Current: 1/1
• CPC Class Codes: G06F 16/2465 Query processing support for facilitating data mining operations in structured databases
• CPC Class Codes: G06F 18/295 Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
• CPC Class Codes: G06F 21/6245 Protecting personal data, e.g. for financial or medical purposes
• CPC Class Codes: G06V 40/53 Measures to keep reference information secret, e.g. cancellable biometrics
• CPC Class Codes: H04L 67/02 based on web technology, e.g. hypertext transfer protocol [HTTP]

Abstract

A method classifies data to determine hidden states of a system, by first randomly permuting the data and inserting client to generate private data. A server classifies the private data according to a hidden Markov model (HMM) to obtain permuted noisy estimates of the states and the chaff, which are returned to the client. The client then removes the chaff, inverts the permuted noisy estimates to obtain unpermuted noisy estimates of the states.

Claims

1. A method for classifying data to determine hidden states of a system, comprising:

randomly permuting the data, acquired from the system by a client, to generate permuted data;

inserting, by the client, chaff in the permuted data to generate private data;

transmitting, by the client to the server, the private data;

classifying, by the server, each sample of the private data independently according, to a hidden Markov model (HMM) to obtain permuted noisy estimates of the states and the chaff;

returning, by the server to the client, the permuted noisy estimates of the states and the chaff;

removing, by the client, the chaff;

inverting, by the client after removing the chaff, the permuted noisy estimates to obtain unpermuted noisy estimates of the states; and

correcting errors, by the client, to obtain estimates of the hidden states.

2. The method of claim 1, wherein server has either partial knowledge of a statistical distribution of the private data given each possible state or a trained classifier that estimates the states, and the client has partial knowledge of a state transition distribution.

3. The method of claim 1, wherein parameters of the HMM include distributions of state transition statistics and statistics of the private data.

4. The method of claim 1, wherein the chaff is random data.

5. The method of claim 1, wherein the correcting uses temporal correlations to obtain the estimates of the hidden states.