Ryo Kikuchi

Koki Hamada, Dai Ikarashi, Ryo Kikuchi et al.

We propose a secure multi-party computation (MPC) protocol that constructs a secret-shared decision tree for a given secret-shared dataset. The previous MPC-based decision tree training protocol (Abspoel et al. 2021) requires $O(2^hmn\log n)$ comparisons, being exponential in the tree height $h$ and with $n$ and $m$ being the number of rows and that of attributes in the dataset, respectively. The cause of the exponential number of comparisons in $h$ is that the decision tree training algorithm is based on the divide-and-conquer paradigm, where dummy rows are added after each split in order to hide the number of rows in the dataset. We resolve this issue via secure data structure that enables us to compute an aggregate value for every group while hiding the grouping information. By using this data structure, we can train a decision tree without adding dummy rows while hiding the size of the intermediate data. We specifically describes a decision tree training protocol that requires only $O(hmn\log n)$ comparisons when the input attributes are continuous and the output attribute is binary. Note that the order is now \emph{linear} in the tree height $h$. To demonstrate the practicality of our protocol, we implement it in an MPC framework based on a three-party secret sharing scheme. Our implementation results show that our protocol trains a decision tree with a height of 5 in 33 seconds for a dataset of 100,000 rows and 10 attributes.

Nuttapong Attrapadung, Koki Hamada, Dai Ikarashi et al.

Privacy-preserving machine learning (PPML) aims at enabling machine learning (ML) algorithms to be used on sensitive data. We contribute to this line of research by proposing a framework that allows efficient and secure evaluation of full-fledged state-of-the-art ML algorithms via secure multi-party computation (MPC). This is in contrast to most prior works, which substitute ML algorithms with approximated "MPC-friendly" variants. A drawback of the latter approach is that fine-tuning of the combined ML and MPC algorithms is required, which might lead to less efficient algorithms or inferior quality ML. This is an issue for secure deep neural networks (DNN) training in particular, as this involves arithmetic algorithms thought to be "MPC-unfriendly", namely, integer division, exponentiation, inversion, and square root. In this work, we propose secure and efficient protocols for the above seemingly MPC-unfriendly computations. Our protocols are three-party protocols in the honest-majority setting, and we propose both passively secure and actively secure with abort variants. A notable feature of our protocols is that they simultaneously provide high accuracy and efficiency. This framework enables us to efficiently and securely compute modern ML algorithms such as Adam and the softmax function "as is", without resorting to approximations. As a result, we obtain secure DNN training that outperforms state-of-the-art three-party systems; our full training is up to 6.7 times faster than just the online phase of the recently proposed FALCON@PETS'21 on a standard benchmark network. We further perform measurements on real-world DNNs, AlexNet and VGG16. The performance of our framework is up to a factor of about 12-14 faster for AlexNet and 46-48 faster for VGG16 to achieve an accuracy of 70% and 75%, respectively, when compared to FALCON.