3 accepted papers
Adversarial vulnerabilities in vision-language models pose a critical challenge to the reliability of large language systems, where typographic manipulations and adversarial perturbations can effectively bypass language model defenses. We introduce Attack as Defense (AsD), the first approach to proa
In the treatment of epilepsy with intracranial electroencephalogram(iEEG), the recognition accuracy is low, and it is difficult to find the correlation between channels because of the large amount of channel numbers and time series data. In order to solve these problems, we propose a novel EEG featu…
Automatic emotion recognition based on electroencephalo-graphic (EEG) signals has received increasing attention in recent years. The Deep Residual Networks (ResNets) can solve vanishing gradient problem and exploding gradient problem well in computer vision and can learn more profound semantic infor…