The repository contains an attempt to reproduce the results published in the article Privacy-Preserving Visual Learning Using Doubly Permuted Homomorphic Encryption.

Contents

• Federated learning
• Problem description
• Solutions
• Experimental results
• Usage and examples
• References

Federated learning

Standard machine learning approaches require centralizing the training data on one machine or in a datacenter. For models trained from user interaction with mobile devices or IoTs, an additional approach was introduced: Federated Learning.

Federated Learning enables devices(computers, mobiles, IoTs) to collaboratively learn a shared prediction model while keeping all the training data on device, decoupling the ability to do machine learning from the need to store the data in the cloud. This goes beyond the use of local models that make predictions on smart devices by bringing model training to the device as well.

It works like this: your device downloads the current model, improves it by learning from data on your phone, and then summarizes the changes as a small focused update. Only this update to the model is sent to the cloud, using encrypted communication, where it is immediately averaged with other user updates to improve the shared model. All the training data remains on your device, and no individual updates are stored in the cloud.

Federated Learning advantages:

• Ability to learn from multiple users while keeping their privacy.
• No need to store extra data in cloud.
• Less network communication.

Problem description

Solutions

Experimental results

1. Reference values

References/Dataset	Caltech101 accuracy	Caltech256 accuracy
Article data	89.3%	74.7%
Simple linear layer	89.0%	76.7%

2. Unencrypted distributed network

Distributed network run 100 epochs per users, and 50 epochs per aggregator.

Workers/Dataset	Caltech101 accuracy/duration	Caltech256 accuracy/duration
5 workers	89.6% / 220s	77.1% / 1117s
10 workers	88.9% / 213s	77.2% / 1130s
50 workers	87.7% / 310s	74.4% / 1296s
100 workers	84.0% / 523s	66.7% / 1381s

3. Fully encrypted distributed network

Workers/Dataset	Caltech101 accuracy/duration	Caltech256 accuracy/duration
5 workers	83% / 3666ss	72% / 9747s
10 workers	81% / 6936s	68.3% / 17954s

4. Encrypted network with sparse updates(keep top 1% deltas and encrypt them)

Workers/Dataset	Caltech101 accuracy/duration	Caltech256 accuracy/duration
5 workers	80.7% / 600s	64,7% / 2057s
10 workers	77.6% / 805s	63.9% / 2594s
50 workers	74% / 2754s	55.0% / 7352s
100 workers	71% / 5372s	43.2% / 13144s

5. Encrypted network with sparse updates(directly learn over delta and use elastic_nets for constraining them)

Workers/Dataset	Optimizer	Caltech101 accuracy/duration	Caltech256 accuracy/duration
5 workers	SGD	83.7% / 606s	68.1% / 2137s
5 workers	Adam	86.2% / 612s	74.1% / 2260s

Usage and examples

References

1. Ryo Yonetani, Vishnu Naresh Boddeti, Yoichi Sato: Privacy-Preserving Visual Learning Using Doubly Permuted Homomorphic Encryption. Published in IEEE International Conference on Computer Vision. doi: 10.1109/ICCV.2017.225

2. Federated Learning