We design a decentralized stochastic optimization algorithm for machine-to-machine (M2M) networks, in which machines in the networks can attend the optimization process in a randomly-unexpected manner. This property of random attendance aims at capturing unexpected events that prevent the machines to successfully communicate with each other. The property, beside simulating the unforeseen nature of the networks, additionally helps reduce the communication cost of machines while still maintaining a reasonable optimization outcome. Our contribution gives practical insights for deploying optimization algorithms over machines in complex environments, including the Internet of Things.

This paper considers the scenario that multiple data owners wish to apply a machine learning method over the combined dataset of all owners to obtain the best possible learning output but do not want to share the local datasets owing to privacy concerns. We design systems for the scenario that the stochastic gradient descent (SGD) algorithm is used as the machine learning method, because SGD (or its variants) is at the heart of recent deep learning techniques over neural networks. Our systems differ from the existing systems in the following features: 1) any activation function can be used, meaning that no privacy-preserving-friendly approximation is required; 2) gradients computed by SGD are not shared but the weight parameters are shared instead; and 3) robustness against colluding parties even in the extreme case that only one honest party exists. One of our systems requires a shared symmetric key among the data owners (trainers) to ensure the secrecy of the weight parameters against a central server. We prove that our systems, while privacy preserving, achieve the same learning accuracy as SGD and, hence, retain the merit of deep learning with respect to accuracy. Finally, we conduct several experiments using benchmark datasets and show that our systems outperform the previous system in terms of learning accuracies.

We present a privacy-preserving deep learning system in which many learning participants perform neural network-based deep learning over a combined dataset of all, without revealing the participants' local data to a central server. To that end, we revisit the previous work by Shokri and Shmatikov (ACM CCS 2015) and show that, with their method, local data information may be leaked to an honest-but-curious server. We then fix that problem by building an enhanced system with the following properties: 1) no information is leaked to the server and 2) accuracy is kept intact, compared with that of the ordinary deep learning system also over the combined dataset. Our system bridges deep learning and cryptography: we utilize asynchronous stochastic gradient descent as applied to neural networks, in combination with additively homomorphic encryption. We show that our usage of encryption adds tolerable overhead to the ordinary deep learning system.

Logistic regression is a powerful machine learning tool to classify data. When dealing with sensitive data such as private or medical information, cares are necessary. In this paper, we propose a secure system for protecting the training data in logistic regression via homomorphic encryption. Perhaps surprisingly, despite the non-polynomial tasks of training in logistic regression, we show that only additively homomorphic encryption is needed to build our system. Our system is secure and scalable with the dataset size.

In this paper, we first present the notion of stateful identity-based encryption (IBE) and then extend standard security definitions for IBE to the stateful setting. After that, we demonstrate a concrete stateful IBE scheme, whose security meets the strongest definition in the setting in random oracle model, and whose encryption and decryption are very efficient, compared to existing IBEs: one pairing each for encryption and decryption.