Efficient Logistic Regression on Large Encrypted Data

Re:Efficient Logistic Regression on Large Encrypted Data by testtest7777 on Wednesday July 11, 2018 @03:51PM (#17509792)

The authors should do a comparison with recent work on HE for inference on other model types. Logistic regression is a subset of neural network inference, and there is significant prior work that appears to be competitive with their results.

Recent work [0] shows that MNIST, for example, can be done in a model secure/data secure way with online 30ms inference latency. For ~500,000 samples, evaluated sequentially, that's roughly 5 hours, which beats their time by about 3x. Perhaps I'm missing details in my read of the paper.

[0] Gazelle: A Low Latency Framework for Secure Neural Network Inference https://arxiv.org/abs/1801.05507

[ Parent | on HN ]

Re:Efficient Logistic Regression on Large Encrypted Data by ssmiler on Wednesday July 11, 2018 @04:30PM (#17510054)

Gazelle paper uses an interactive protocol ("combination of homomorphic encryption and traditional two-party computation techniques") compared to thread paper.

[ Parent | on HN ]

Re:Efficient Logistic Regression on Large Encrypted Data by TTPrograms on Wednesday July 11, 2018 @05:03PM (#17510261)

It's also just inference, not training.

[ Parent | on HN ]

Re:Efficient Logistic Regression on Large Encrypted Data by confounded on Wednesday July 11, 2018 @09:42PM (#17511615)

> inference latency... that's roughly 5 hours,

The 17 hours in the article is to derive the 'encrypted' model. Training takes significantly longer than prediction.

[ Parent | on HN ]