- Keywords: Neuro-AI interface, Neuroscore, Generative adversarial networks, EEG, Convolutional neural networks
- TL;DR: Describe a neuro-AI interface technique to evaluate generative adversarial networks
- Abstract: Deep neural networks (DNNs) are inspired from the human brain and the interconnection between the two has been widely studied in the literature. However, it is still an open question whether DNNs are able to make decisions like the brain. Previous work has demonstrated that DNNs, trained by matching the neural responses from inferior temporal (IT) cortex in monkey's brain, is able to achieve human-level performance on the image object recognition tasks. This indicates that neural dynamics can provide informative knowledge to help DNNs accomplish specific tasks. In this paper, we introduce the concept of a neuro-AI interface, which aims to use human's neural responses as supervised information for helping AI systems solve a task that is difficult when using traditional machine learning strategies. In order to deliver the idea of neuro-AI interfaces, we focus on deploying it to one of the fundamental problems in generative adversarial networks (GANs): designing a proper evaluation metric to evaluate the quality of images produced by GANs.