No abstract uploaded!

No abstract uploaded!

In this paper, we introduce a new provably secure ephemeral-only RLWE+Rounding-based key exchange protocol and a proper approach to more accurately estimate the security level of the RLWE problem with only one sample. Since our scheme is an ephemeral-only key exchange, it generates only one RLWE sample from protocol execution. We carefully analyze how to estimate the practical security of the RLWE problem with only one sample, which we call the ONE-sample RLWE problem. Our approach is different from existing approaches that are based on estimation with multiple RLWE samples. Though our analysis is based on some recently developed techniques in Darmstadt, our type of practical security estimate was never done before and it produces security estimates substantial different from the estimates before based on multiple RLWE samples. We show that the new design improves the security and reduce the communication cost of the protocol simultaneously by using one RLWE+Rounding sample technique. We also present two parameter choices ensuring 2^{−60} key exchange failure probability which cover security of AES-128/192/256 with concrete security analysis and implementation. We believe that our construction is secure, simple, efficient and elegant with wide application prospects.

It is well acknowledged in image processing domain that the information can be decomposed into different frequency parts and each part has its own merits. However, existing neural networks always ignore the distinctions and straightforwardly feed all the information into neural networks together, treating them equally. In this paper, we propose a novel neural networks framework named J-Net that decomposes images into different frequency bands and then processes them sequentially. Concretely, the images have been decomposed by wavelet transformation and then the wavelet coefficients are fed into neural networks gradually in different depth according to their decomposition levels. An attention module is utilized to facilitate the fusion of neural network features and injected information, yielding significant performance gain. Furthermore, we show how does the information with different frequency impact the accuracy of neural networks. Experiments show that 5.91%, 5.32% and 2.00% accuracy improvements on Caltech 101, Caltech256 and ImageNet, respectively.

No abstract uploaded!