# US 2021/0174243 A1 · Efficient private vertical federated learning

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0174243 A1
- Document ID: US20210174243A1
- Status: Published application
- Kind: A1
- Application number: 16/706,328
- Filed: 2019-12-06
- Published: 2021-06-10
- Earliest claimed: 2019-12-06
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Nathalie Baracaldo Angel; Runhua Xu; Yi Zhou; Ali Anwar; Heiko H. Ludwig
- Web page: https://patentgrep.com/patent/US20210174243A1

## CPC

- H04L 9/08: Key distribution {or management, e.g. generation, sharing or updating, of cryptographic keys or passwords (network architectures or network communication protocols for supporting key management in a packet data network H04L63/06)}
- H04L 9/083: {involving central third party, e.g. key distribution center [KDC] or trusted third party [TTP]}
- G06F 16/903: Querying (for retrieval from the web G06F16/953)
- G06N 20/00: Machine learning
- G06N 20/20: Ensemble learning
- H04L 9/0825: {using asymmetric-key encryption or public key infrastructure [PKI], e.g. key signature or public key certificates}
- H04L 9/30: Public key, i.e. encryption algorithm being computationally infeasible to invert or user's encryption keys not requiring secrecy
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Systems and techniques that facilitate universal and efficient privacy-preserving vertical federated learning are provided. In various embodiments, a key distribution component can distribute respective feature-dimension public keys and respective sample-dimension public keys to respective participants in a vertical federated learning framework governed by a coordinator, wherein the respective participants can send to the coordinator respective local model updates encrypted by the respective feature-dimension public keys and respective local datasets encrypted by the respective sample-dimension public keys. In various embodiments, an inference prevention component can verify a participant-related weight vector generated by the coordinator, based on which the key distribution component can distribute to the coordinator a functional feature-dimension secret key that can aggregate the encrypted respective local model updates into a sample-related weight vector. In various embodiments, the inference prevention component can verify the sample-related weight vector, based on which the key distribution component can distribute to the coordinator a functional sample-dimension secret key that can aggregate the encrypted respective local datasets into an update value for a global model.

## Claims

### Claim 1

A system, comprising:
a memory that stores computer-executable components; and
a processor, operably coupled to the memory, that executes the computer-executable components stored in the memory, wherein the computer-executable components comprise:
a key distribution component that distributes functional encryption public keys to respective participants in a vertical federated learning framework governed by a coordinator; and
an inference prevention component that verifies query vectors generated by the coordinator, based on which the key distribution component distributes functional encryption secret keys to the coordinator.

### Claim 2 (depends on 1)

The system of claim 1, wherein:
the key distribution component distributes respective feature-dimension public keys and respective sample-dimension public keys to the respective participants, wherein the respective participants send to the coordinator respective local model updates encrypted by the respective feature-dimension public keys and respective local datasets encrypted by the respective sample-dimension public keys;
the inference prevention component verifies a participant-related weight vector generated by the coordinator, based on which the key distribution component distributes to the coordinator a functional feature-dimension secret key that aggregates the encrypted respective local model updates into a sample-related weight vector; and
the inference prevention component verifies the sample-related weight vector, based on which the key distribution component distributes to the coordinator a functional sample-dimension secret key that aggregates the encrypted respective local datasets into an update value for a global model.

### Claim 3 (depends on 2)

The system of claim 2, wherein:
the key distribution component generates the respective feature-dimension public keys and the functional feature-dimension secret key via multi-input functional encryption techniques for inner-products, and generates the respective sample-dimension public keys and the functional sample-dimension secret key via single-input functional encryption techniques for inner-products.

### Claim 4 (depends on 3)

The system of claim 3, wherein:
the inference prevention component verifies a query vector by determining that a number of non-zero elements of the query vector is above a predetermined threshold, and by determining that a number of training epochs is less than both a maximum number of features and a maximum sample batch size established by the coordinator before beginning a training session.

### Claim 5 (depends on 2)

The system of claim 2, wherein the respective participants perform one-way communication with the coordinator, and wherein the respective participants do not communicate with each other.

### Claim 6 (depends on 2)

The system of claim 2, wherein only one of the respective participants possesses labels which the global model is trained to predict.

### Claim 7 (depends on 2)

The system of claim 2, wherein a number of the respective participants changes between consecutive training iterations of a training session.

### Claim 8 (depends on 2)

The system of claim 2, wherein the inference prevention component sends an alert to the respective participants if the participant-related weight vector or the sample-related weight vector is invalid.

### Claim 9

A computer-implemented method, comprising:
distributing, by a device operatively coupled to a processor, functional encryption public keys to respective participants in a vertical federated learning framework governed by a coordinator; and
verifying, by the device, query vectors generated by the coordinator, and accordingly distributing functional encryption secret keys to the coordinator.

### Claim 10 (depends on 9)

The computer-implemented method of claim 9, further comprising:
distributing, by the device, respective feature-dimension public keys and respective sample-dimension public keys to the respective participants, wherein the respective participants send to the coordinator respective local model updates encrypted by the respective feature-dimension public keys and respective local datasets encrypted by the respective sample-dimension public keys;
verifying, by the device, a participant-related weight vector generated by the coordinator, and accordingly distributing to the coordinator a functional feature-dimension secret key that aggregates the encrypted respective local model updates into a sample-related weight vector; and
verifying, by the device, the sample-related weight vector, and accordingly distributing to the coordinator a functional sample-dimension secret key that aggregates the encrypted respective local datasets into an update value for a global model.

### Claim 11 (depends on 10)

The computer-implemented method of claim 10, further comprising:
generating, by the device, the respective feature-dimension public keys and the functional feature-dimension secret key via multi-input functional encryption techniques for inner-products; and
generating, by the device, the respective sample-dimension public keys and the functional sample-dimension secret key via single-input functional encryption techniques for inner-products.

### Claim 12 (depends on 11)

The computer-implemented method of claim 11, wherein the verifying query vectors comprises:
determining, by the device, that a number of non-zero elements of a query vector is above a predetermined threshold; and
determining, by the device, that a number of training epochs is less than both a maximum number of features and a maximum sample batch size established by the coordinator before beginning a training session.

### Claim 13 (depends on 10)

The computer-implemented method of claim 10, wherein the respective participants perform one-way communication with the coordinator, and wherein the respective participants do not communicate with each other.

### Claim 14 (depends on 10)

The computer-implemented method of claim 10, wherein only one of the respective participants possesses labels which the global model is trained to predict.

### Claim 15 (depends on 10)

The computer-implemented method of claim 10, wherein a number of the respective participants changes between consecutive training iterations of a training session.

### Claim 16 (depends on 10)

The computer-implemented method of claim 10, further comprising:
sending, by the device, an alert to the respective participants if the participant-related weight vector or the sample-related weight vector is invalid.

### Claim 17

A computer program product for facilitating universal and efficient privacy-preserving vertical federated learning, the computer program product comprising a computer readable memory having program instructions embodied therewith, the program instructions executable by a processing component to cause the processing component to:
distribute functional encryption public keys to respective participants in a vertical federated learning framework governed by a coordinator; and
verify query vectors generated by the coordinator, and accordingly distribute functional encryption secret keys to the coordinator.

### Claim 18 (depends on 17)

The computer program product of claim 17, wherein the program instructions are further executable to cause the processing component to:
distribute respective feature-dimension public keys and respective sample-dimension public keys to the respective participants, wherein the respective participants send to the coordinator respective local model updates encrypted by the respective feature-dimension public keys and respective local datasets encrypted by the respective sample-dimension public keys;
verify a participant-related weight vector generated by the coordinator, and accordingly distributing to the coordinator a functional feature-dimension secret key that aggregates the encrypted respective local model updates into a sample-related weight vector; and
verify the sample-related weight vector, and accordingly distribute to the coordinator a functional sample-dimension secret key that aggregates the encrypted respective local datasets into an update value for a global model.

### Claim 19 (depends on 18)

The computer program product of claim 18, wherein the program instructions are further executable to cause the processing component to:
generate the respective feature-dimension public keys and the functional feature-dimension secret key via multi-input functional encryption techniques for inner-products; and
generate the respective sample-dimension public keys and the functional sample-dimension secret key via single-input functional encryption techniques for inner-products.

### Claim 20 (depends on 19)

The computer program product of claim 19, wherein the processing component verifies query vectors by:
determining that a number of non-zero elements of a query vector is above a predetermined threshold; and
determining that a number of training epochs is less than both a maximum number of features and a maximum sample batch size established by the coordinator before beginning a training session.
