Hangzhou Hikvision Digital Technology Co., Ltd.

Kezhang Lin

Bin Wang

Si Chen

Wei Wang

Chao Li

Jiadong Chen

Xing Wang

Li Wan

Zhiwei Zhang

Xiaohong Guan

Haonan Yan

Yulong Shen

Weikang Rao

Guoyun Wang

Dechang Zhu

Chaohao Li

Song Liu

Chengrun He

Junwei Ni

A systematic security protection method for a multimodal AI model is provided. The method comprises: in a security detection stage of the target multimodal AI model, according to a perceptual distance between detection data and original data determined by a trained distance-aware judgment model, obtaining detection data whose perceptual distance from the original data meets a requirement, and detecting the target multimodal AI model by using the detection data; in a case that the detection result in the detection stage is that the target multimodal AI model has a security problem, on one hand, performing security coding protection and tokenization filtering on the input of the model, on the other hand, performing multi-dimensional omnidirectional evaluation on the multimodal AI model.

Systematic security protection method for multimodal AI model, apparatus and device

The present disclosure provides a neural network-based security defense method for encrypted multimedia data, including: adding implicit digital watermark information to original encrypted multimedia traffic to obtain watermark encrypted multimedia traffic; determining a first feature sequence of the watermark encrypted multimedia traffic; based on the first feature sequence and a second feature sequence of the original encrypted multimedia traffic, determining a correlation coefficient between the first feature sequence and the second feature sequence; and if the correlation coefficient indicates that a negative correlation between the first feature sequence and the second feature sequence satisfies a preset condition, performing training based on the first feature sequence and the second feature sequence to obtain a target neural network, where the target neural network is configured to identify whether there is implicit digital watermark information in encrypted multimedia traffic to be detected.

Neural network-based security defense method for encrypted multimedia data, electronic device, and computer program product

A data privacy protection method based on multi-layer encryption is disclosed, which is related to the technical field of network technologies. The method includes: by an initial client, encrypting service data based on a public key set to obtain first to-be-processed data, and sending the first to-be-processed data to a next client; by an intermediate client, obtaining the first to-be-processed data sent by a previous client, encrypting service data based on a public key set to obtain second encrypted data, and sending second to-be-processed data and the second encrypted data to a next client; by a final client, obtaining the second to-be-processed data and the second encrypted data sent by a previous client, decrypting the second to-be-processed data and the second encrypted data based on its own private key, to obtain service data of each of other clients other than the final client itself, and sending the service data of all of the clients in a data privacy protection system to a preset service platform. The embodiments of the present application are applied to reduce the risk of user's privacy data leakage.

Data privacy protection method based on multi-layer encryption

Growing community of inventors

Hangzhou, China

Kezhang Lin