Speaker: Kazuto Fukuchi
AIP AI Security & Privacy Team
Recent Advances of Differential Privacy under Local model
Techniques for preserving differential privacy have been incorporated into many real-world systems, including services provided by Apple, Google, and Uber. In particular, most of these systems employ the local model, in which sensitive data must be sanitized before submitting it to the system and ensure the differential privacy under the model. In this talk, I will introduce some recent advances in the field of differential privacy under the local model. The talk mainly focuses on introducing some techniques for analyzing an information theoretic limit of the differentially private mechanism under the local model. These techniques are useful for proving the minimax optimality of some differentially private mechanisms.
Speaker: Tatsuro Kawamoto
National Institute of Advanced Industrial Science and Technology,
The Artificial Intelligence Research Center
Probabilistic Modeling Research Team
Mean-field theory of graph neural networks in graph partitioning
A theoretical performance analysis of the graph neural network (GNN) is presented. For classification tasks, the neural network approach has the advantage in terms of flexibility that it can be employed in a data-driven manner, whereas Bayesian inference requires the assumption of a specific model. A fundamental question is then whether GNN has a high accuracy in addition to this flexibility. Moreover, whether the achieved performance is predominately a result of the backpropagation or the architecture itself is a matter of considerable interest. To gain a better insight into these questions, a mean-field theory of a minimal GNN architecture is developed for the graph partitioning problem. This demonstrates a good agreement with numerical experiments.
|Date||July 26, 2019 (Fri) 13:00 - 16:30|